Product Competitive Analysis Model Based on Consumer Preference Satisfaction Similarity: Case Study of Smartphone UGC

Abstract

Accurately identifying key competitors across multiple product lines is essential for enhancing the flexibility and competitiveness of product strategies. This study introduces a novel data-driven model for competitive analysis termed the Product Competition Analysis Model based on Consumer Preference Satisfaction Similarity (PCAM-CPSS). Unlike traditional methods that rely on assessments of the competitive environment, the PCAM-CPSS leverages sentiment analysis of user-generated content (UGC) to quantify consumer preference satisfaction. This method constructs a network based on product satisfaction similarity to map competitive relationships and employs a community detection algorithm to identify key competitors. To assess the model’s efficacy, we collected and analyzed user reviews of various smartphone brands to serve as an evaluation dataset. We compared the performance of the PCAM-CPSS against two mainstream competitive analysis methods: attribute similarity-based ratings and co-occurrence statistics. The results, evaluated using the Normalized Discounted Cumulative Gain (NDCG) index, demonstrate that the PCAM-CPSS, particularly with price adjustment, offers significant advantages in identifying competitors more accurately than other evaluated methods.

1. Introduction

Product competitive analysis stands as a foundational pillar within the realm of strategic management, empowering enterprises to grasp product competitiveness, discern consumer preferences, and customize products to align with diverse needs [1,2]. In today’s fiercely competitive business landscape, it is imperative for businesses to swiftly and accurately gauge products’ positions [3,4]. Traditional competitive analysis [5,6,7] primarily focuses on executive perspectives, emphasizing competitor identification, but tends to be overly macroscopic, static, and qualitative, making it less adaptable to the rapidly evolving digital age. It is crucial to recognize that products ultimately serve consumers, and their purchasing decisions are influenced by consumer perceptions and responses [8,9]. In reality, products satisfying similar consumer needs engage in true competitive dynamics. Therefore, prioritizing consumer preference adds more value compared to the traditional product competitive analysis approach.

The rapid emergence of information and communication technologies, such as e-commerce, social media, and online forums, has led to the widespread dissemination of UGC, which has garnered significant research on user-centered analysis [10,11,12,13]. UGC is particularly noteworthy for its distinct advantages over traditional product competitive analysis data, including diverse sources, extensive scale, low cost, real-time availability, and easy accessibility [14,15]. These attributes not only offer UGC as a more direct and authentic resource for competitive analysis but also enhance the speed of analysis and the adaptability of businesses to market changes [16,17]. While UGC utilization may encounter challenges like handling extensive data, varying data quality, content diversity, and sentiment analysis, the rapid advancements in computer technology provide effective solutions. Machine learning and deep learning techniques automate textual analysis, while social network analysis and natural language processing uncover crucial relationships and trends in UGC [18]. For instance, Transformer-based language models have improved the accuracy of nearly all NLP tasks, providing rich opportunities for UGC-based product innovation [19]. Therefore, the integration of advanced computer technology with UGC in competitive analysis methods has become an indispensable tool for meeting market needs and addressing market competition [20,21].

Current research in product competitive analysis based on UGC primarily relies on co-occurrence pattern mining [22,23,24], product attribute similarity [14,25,26], and consumer consideration sets [27,28,29]. Nevertheless, the use of co-occurrence pattern mining is often plagued by biases due to the intricate nature of semantic rule configuration and the scarcity of co-occurrence relationships [30]. In contrast, competition identification based on product attribute similarity offers a solution to this challenge. However, to enhance the precision of competition identification, it necessitates intricate natural language processing to extract the unique attributes of each product. Nonetheless, the primary limitation of this approach lies in its potential oversight of competitors with significantly different attributes [14]. Although competition identification grounded in consumer consideration sets alleviates some issues, companies face formidable challenges in accessing consumer shopping behavior data concerning their competitors.

This research provides a wide array of perspectives yet confronts formidable technical and data procurement impediments. Drawing on product attribute similarity analysis, we introduce a novel Product Competition Analysis Model grounded in Consumer Preference Satisfaction Similarity (CPSS). A notable advantage of this newly devised model is its capacity to leverage a more accessible corpus of post-purchase reviews. By employing consumer reviews, it expands the selection of competitors across a broader spectrum. Eventually, the method introduced in this article has been established to exhibit superior accuracy in both competitor identification and assessing the global competitiveness of products when contrasted against two prominent UGC-based product competitive analysis methodologies [23,31]. The correlated research findings harbor potential within the domain of competitive analysis in business operations.

The innovations in this research can be summarized as follows: first, it employs sentiment analysis at the product attribute level, accounting for pricing factors in consumer decision-making, and comprehensively evaluates the extent to which products fulfill consumer preference, overcoming challenges in existing UGC-based competitive analysis models related to complex semantic rules, limited co-occurrence patterns, and insufficient identification of key factors. Second, a product competitive relationship network is constructed, which no longer relies solely on product attribute similarity but is grounded in the similarity of products with respect to overall consumer preference satisfaction. Third, through community discovery and social network analysis, this study effectively identifies the competitive market where the target product operates and evaluates its global competitiveness.

The rest of this paper is structured as follows: Section 2 presents the literature review; Section 3 describes the PCAM-CPSS; Section 4 validates the model through case data; Section 5 summarizes the research findings; Section 6 discusses the practical and theoretical contributions; and Section 7 addresses the limitations and suggests for future work.

2. Literature Review

2.1. Product Competitive Analysis

Globalization and digitalization in the market lead to information overload for businesses, making competitive analysis more challenging [32]. In addition to existing competitors in the market, new competitors often enter through gaps in underserved areas such as low-end markets or emerging markets [33]. This necessitates incumbents to gather more market data and expand their identification scope to discover potential competitors [14], but this significantly increases the workload of competitor assessment. Therefore, designing automated competitive analysis solutions to enhance analysis accuracy is a research area that deserves further development [25,34]. Product competitive analysis typically involves two key steps: identification and evaluation. Firstly, in the process of identifying competitors, existing research often utilizes various data-mining tools to recognize competitors, which combine relation extraction and entity recognition [35]. By analyzing data from patents [36,37], web search [38], and social media [14,39], these tools help discover existing, new, or potential competitors. Subsequently, competitors’ products are assessed according to established evaluation criteria. A multidimensional set of indicators is generally employed to characterize competitors’ competitive levels, and a comprehensive analysis of their strengths, advantages, and weaknesses is conducted by weighing these indicators’ scores and rankings [27,40]. It is worth noting that product competitive analysis is not a one-time activity but should be considered an essential component of a company’s ongoing assessment and strategic adjustment. Companies need to update their analysis data regularly to ensure they stay informed of the latest competitive landscape.

2.2. Consumer Preference Satisfaction

In order to clarify the concept of consumer preference satisfaction, it is important to have a fundamental understanding of consumer satisfaction [13]. Building upon the literature [41], the definition of consumer satisfaction within specific contexts typically comprises three fundamental conditions. First, it embodies a varied affective response, encompassing a range of emotional and cognitive reactions with differing intensities. Second, it is time-specific, subject to changes over time, and characterized by emotional responses that fluctuate temporally. Third, consumer satisfaction is directed towards critical aspects of product acquisition and consumption. Consumer preference satisfaction serves as a more specific subset of consumer satisfaction, emphasizing the extent to which a product or service fulfills the actual needs of consumers. This form of satisfaction is manifested through the disparity between specific expectations and experiential perceptions before and after consumption [42], often culminating in an emotional reflection formed through a comprehensive evaluation of product performance [43], service [44], and other pertinent factors [45]. Hence, consumer preference satisfaction assumes a pivotal role as an intermediary for evaluating the concordance between consumer expectations and the efficacy of products or services. Through a systematic examination of consumer satisfaction concerning product attributes and experiential dimensions, it affords the opportunity to cultivate a more discerning and nuanced understanding of the relative competitive positioning of diverse products within the market landscape.

2.3. Social Network Analysis

Constructing comparative networks with products as nodes and competitive relationships as edges has been widely employed in product competition analysis. Some research utilizes node centrality as a key indicator for measuring the strength of product competitiveness [46,47]. However, this approach solely considers the influence of neighboring nodes while overlooking the relationship between the target node and non-neighboring nodes, thus failing to accurately assess the product’s position in the entire network. To elucidate complex competitive connections among various products, researchers have sought to employ a multitude of network analysis methodologies. Netzer et al. utilized co-occurrence relationships as the basis for product competitiveness identification, constructing an undirected network to examine the competitiveness of different brands’ products in the market [23]. Moreover, other studies have regarded similar product attributes as the foundation for competitive relationships, connecting distinct products by calculating the similarity of user preferences [25,48]. Notably, with the rapid development of digital technology, numerous contemporary research endeavors apply SNA methods to analyze social media, online reviews, and other UGC for a more comprehensive evaluation of product competitiveness and market conditions [14,22,49].

2.4. Discussion of the Literature

In the dynamic competitive landscape influenced by elevated consumer expectations, this study presents a product competitive analysis model named PCAM-CPSS. This model quantifies consumer preference satisfaction through attribute-level sentiment analysis and UGC-based attribute frequency assessment, enabling the identification of intricate competitive relationships crucial for understanding market dynamics. Advanced techniques like social network analysis reveal the nuanced competitive market landscape. This research evaluates product competitiveness by analyzing distinctions in the PCAM-CPSS, supported by a directed weighted network and an importance-ranking algorithm. Empirical validation confirms the superiority of the PCAM-CPSS model in product competitor identification and overall competitiveness ranking. This study contributes a comprehensive understanding of product competition dynamics, showcasing the efficacy of the PCAM-CPSS model in navigating the complexities of the modern competitive market.

3. Model Description

This research focuses on the consumer’s selection and evaluation processes during purchase decisions, wherein each consumer endeavors to acquire the product that maximizes personal satisfaction. To establish a competitive intelligence analysis model for products, we introduce the PCAM-CPSS. The framework of the PCAM-CPSS is depicted in Figure 1, comprising four modules. Through the analysis of target products’ competition, we constructed a competitive relationship network that includes the target products and their potential competitors, providing valuable insights into product positioning and market strategies.

3.1. Consumer Preference Satisfaction Framework

Consumer preference satisfaction can be characterized as the comprehensive satisfaction derived from a product, encompassing all assessable attributes pertaining to the consumer’s expectations and preferences. As a result, its measurement involves assessing the performance ($e_j$) of various product attributes and the level of importance ($d_i$) that consumers attach to these attributes, as depicted in Figure 2. In this representation, solid lines between consumer groups and products indicate that the product meets the needs of that particular consumer group, whereas dashed lines signify that the product fails to meet the needs of that consumer group. Recognizing that product purchase decisions are contingent upon consumers’ perceptions and responses to different products, the authenticity and accuracy of consumer sentiments toward the product, as reflected in UGC, prove invaluable. These sentiments, along with the frequency of mentions by consumers, are employed to gauge the performance of various product attributes and the significance consumers attach to these attributes. Therefore, this study hypothesizes that the extent to which a product fulfills consumer preference is assessed by the product of these two measurements.

3.1.1. Competitive Products Preliminary Selection

Consumer choices are subject to the dual influence of preferences and budgetary constraints. In the process of making purchase decisions, consumers embark on an initial screening process, encompassing products that align with financial considerations, thereby establishing the preliminary consideration set. Consequently, competitive products ought to be provisionally curated based on specific price brackets:

$$P_C=\{p_i\left|p{r}_1\le p{r}_i\le p{r}_2,p_i\in P\}\right.$$

(1)

where $P$ refers to the comparable product set of the target products, $P_C$ refers to the competitive product set selected according to the price range of the products, including the target products, and $p{r}_i$ refers to the price of the product $p_i$. The variables $p{r}_1$ and $p{r}_2$ represent the lower and upper thresholds of selected product prices.

3.1.2. Consumer Performance Measurement

Post-purchase online reviews within UGC serve as the predominant medium through which consumers articulate their opinions and sentiments, assessing the actual attributes of the product. The various attributes of each product can be effectively measured by analyzing consumers’ sentiments ($e_j=\left[e_{j1},e_{j2},\cdots ,e_{jn}\right]$) for various attributes $(p_j=[p_1, p_2,\dots ,p_m\left]\right)$ [50,51]. We define the numerical representation of product characteristics as product attribute performance. The quantitative computation methodology relies on aspect-based sentiment analysis within the framework of consumer reviews. It involves the rigorous identification and evaluation of emotional expressions integrated within the textual descriptions concerning different aspects of the reviewed product [52]. The process of extracting opinion tuples, comprising product attributes, modifiers, and opinion words (e.g., “screen”, “very”, “clear”), facilitates the computation of sentiment scores for individual product attributes, thereby enabling a nuanced aspect-based sentiment analysis of these attributes.

First, text preprocessing. Our analysis began by applying a series of fundamental preprocessing steps, including sentence segmentation, word segmentation, and part-of-speech tagging. This task was efficiently executed through the utilization of the Stanford CoreNLP toolkit, a comprehensive natural language-processing instrument recognized for its extensive functionality, user-friendly interface, rapid processing capabilities, robustness, and multilingual support [53].

Second, dependency parsing. Due to the subsequent phase of our analysis involving syntactic dependency analysis of textual content, it is imperative to establish explicit rules for identifying attribute–viewpoint tuples. This includes the identification of pairs consisting of (product attribute, opinion word) predicated on the sentence’s dependency structure, and the recognition of pairs of (modifier, opinion word) by scrutinizing adverbial–headword relationships within the sentence.

Third, sentiment scoring. Attribute-level sentiment analysis was executed employing SnowNLP, a well-established tool recognized for its adeptness in sentiment analysis. This tool leverages a pre-trained sentiment analysis model founded on Chinese user product evaluation datasets, rendering it an apt choice for the analysis of sentiments expressed within online reviews [22].

Fourth, performance evaluation. A fundamental component of our study involves the establishment of a product attribute dictionary and the evaluation of diverse product attributes. On one hand, the construction of the product attribute lexicon entails the random sampling of attribute terms extracted from individual product reviews, amalgamated with data obtained from the product user manual. On the other hand, manual categorization of product attributes was conducted to facilitate the evaluation process. Sentiment analysis generates scores that quantitatively assess attribute performance.

3.1.3. Consumer Attention Measurement

Consumer groups with similar product needs have varying preferences for product attributes [54]. In essence, consumer cohorts exhibit diminutive clusters characterized by analogous attention spans. For example, photography enthusiasts prioritize camera and memory performance, while business professionals prioritize signal strength and battery endurance in smartphones. It can be argued that consumer groups $c_i$ purchasing the same product $p_i$ tend to have similar preferences for product attributes, indicating a shared need for that product category. Considering that consumers place a greater emphasis on specific product attributes, their tendency to express opinions and sentiments regarding those attributes is stronger. Therefore, the degree of importance that consumer group ($c_i$) places on product attributes $A=\{a\left|a=\mathrm{1,2},\dots ,n\}\right.$ of product $p_i$ can be measured by the frequency of mentions ($d_{ia}$) by $c_i$. The distribution of the frequency of mentions of product attributes reflects the degree of importance that $c_i$ attaches to these attributes, as represented by $d_i=[d_{i1},d_{i2},\dots ,d_{in}]$, where $d_{i1}+d_{i2}+\cdots +d_{in}=1$. Ultimately, the collection of the degree of importance of product attributes by consumer groups possessing similar needs, as measured by the frequency of mentions of product attributes by purchasers of different products, is represented as $D=\left\{d_i\right|i=\mathrm{1,2},\dots ,m\}$.

3.1.4. Consumer Preference Satisfaction Calculation

During the consumer decision-making process, consumers assess different products by considering their attention and perception of product attributes [54]. The weight assigned to each product attribute is determined by consumers’ attention. The overall satisfaction of similar consumer preference with each product is measured by the product of attribute attention vector $d_i$ and attribute performance vector $e_j$:

$$W\left(d_i,e_j\right)=d_i\times {e_j}^T$$

(2)

where the magnitude of $W(d_i,e_j)$ indicates the level at which product $j$ satisfies the needs of consumer group $c_i$.

3.2. Competitive Relationship Network

The product competitive relationship network of CPSS is based on consumer preference satisfaction similarity and incorporates the impact of product price. An undirected weighted network is formed by selectively retaining edges representing relationships of higher strength while filtering out those with lower strength. This approach takes into account comprehensive consumer satisfaction, thereby enabling the precise identification of competitive relationships.

3.2.1. Competitive Relationship Intensity

The competitive relationship among products is determined by the similarity of consumer preference satisfaction and their cumulative values. Higher intensity indicates greater similarity and increased consumer indecision. The intensity is measured by assigning weights based on the number of consumer reviews ($r_i$) for each product. To standardize across various product launch times, the review count was averaged relative to the duration each product has been on sale. A minimum threshold ($\alpha , 0<\alpha <1$) for consumer preference satisfaction was set to avoid very low similarity. The calculation of competitive relationship intensity between two products can be summarized as follows:

$$Cr\left(p_j,p_k\right)=\sum _{i\in [1,m],W(d_i,e_j)\ge \alpha ,W(d_i,e_k)\ge \alpha }\mathit{exp}\left(-\mathit{ln}\left|W\left(d_i,e_j\right)-W\left(d_i,e_k\right)\right|\right)\times {r_i}^{\prime }$$

(3)

$$r_i^{\prime }={\displaystyle \frac{r_i}{{\sum }_{i=1}^m{r}_i}}$$

(4)

where $\left|W(d_i,e_j)-W(d_i,e_k)\right|$ refers to the difference between the product $p_j$ and $p_k$ in satisfaction of consumer need $d_i$. The conditions $W(d_i,e_j)\ge \alpha$ and $W(d_i,e_k)\ge \alpha$ ensure that the competitive relationship intensity calculation considers the difference between the consumer preference satisfaction of products $p_j$ and $p_k$ when both reach the minimum threshold $\alpha$. The larger the $Cr(p_j,p_k)$, the stronger the competitive relationship between products.

3.2.2. Competitive Relationship Weight Adjustment

We consider both consumer preference satisfaction and the impact of product price on the competitive relationship. Higher prices may decrease overall satisfaction but provide more possibilities to meet consumer preference [55]. To capture realistic scenarios, we calculated the weight of the competitive relationship, adjusted for price, by considering overall consumer preference satisfaction. The weight of the competitive relationship, adjusted for price, was calculated as follows:

$$Cr{\left(p_j,p_k\right)}^{\prime }=\sum _{i\in [1,m],W(d_i,e_j)\ge \alpha ,W(d_i,e_k)\ge \alpha }\mathit{exp}\left(-\mathit{ln}\left|{\displaystyle \frac{W\left(d_i,e_j\right)}{p{r}_j}}-{\displaystyle \frac{W\left(d_i,e_k\right)}{p{r}_k}}\right|\right)\times {r_i}^{\prime }$$

(5)

where $p{r}_j$ and $p{r}_k$ are the prices of products $p_j$ and $p_k$, respectively. ${\displaystyle \frac{W(d_i,e_j)}{p{r}_j}}$ refers to the satisfaction of consumer need $d_i$, and of product $p_j$ with the price adjustment.

3.2.3. Product Competition Network Construction

This study constructed the PCAM-CPSS product competitive relationship network based on consumer preference satisfaction and price-adjusted competitive relationship intensity. The competitive product set $P_C$ was screened within a price range, with each product $p_i$ as a node and the number of consumer reviews $r_i$ as the degree of node. Competitive relationships were represented by undirected weighted edges between products, where the edge weight was determined by the price-adjusted competitive relationship intensity ($Cr{(p_j,p_k)}^{\prime }$). To eliminate weak competitive relationships, a minimum threshold $\beta$ was set for the product’s competitive relationship intensity.

3.3. Product Competitor Identification

To identify competitive competitors and closely connected product groups, we employed the Speaker–Listener Label Propagation Algorithm (SLPA) [56] for community detection in the PCAM-CPSS product’s competitive relationship network. Community detection plays a crucial role in understanding the network structure and identifying distinct market segments. The effectiveness of the SLPA algorithm in detecting overlapping communities within networks has been empirically substantiated, rendering it a judicious selection for our analytical framework. Our principal objective revolves around the optimization of community detection, achieved through the deliberate construction of a weighted network structure. Consequently, we introduced specific refinements to the implementation protocol of the SLPA algorithm, engendering the subsequent adaptations:

Step 1: process of node label initialization. Initialize the label information of all product nodes in the PCAM-CPSS product’s competitive relationship network with unique labels. Each node’s memory only contains the initialized labels, and the weight value is set to 1.

Step 2: process of label propagation. (A) Select a product node as the listener. (B) Each neighbor node of the current product node acts as a speaker. The weight value in its memory determines the probability for each label. An automated program randomly selects a label and sends it to the listener. Record the weight of the edge connecting the speaker and the listener. (C) Count the received labels and their weights from the neighboring nodes of the current product. Choose the label with the highest sum of weights as the new label. If the new label already exists in the current node’s memory, increase its weight value by 1. Otherwise, add the new label to the node’s memory with a weight value of 1. (D) Repeat (A) to (C) until the algorithm convergence or ergodic reaches the specified frequency, and the algorithm ends. Otherwise, the tag propagates in the process of continuous ergodic.

Step 3: Proceed with label classification. Remove labels from the product node’s memory if their weight falls below the threshold value. Identify product nodes with the same label as a community for community detection.

3.4. Product Competitiveness Assessment

Based on the community detection results of the PCAM-CPSS product’s competitive relationship network, we performed a comparative analysis to assess the competitiveness of the target product within its specific market segment. A comparative network was constructed to evaluate the overall competitive position of the target product. The ranking of products is based on the significance of their nodes in the network.

3.4.1. Competitive Relationship Identification

Establishing effective competitive relationships between products requires assessing the strength of these relationships. Using the competitive relationship network, we can measure the intensity of competitive relationships between products. An edge between products $p_j$ and $p_k$ in this network signifies a significant and effective competitive relationship. The competitiveness of a product is evaluated based on consumers’ overall assessment, represented by the weighted D-value of consumer preference satisfaction. The calculation is as follows:

$$C_c\left(p_j,p_k\right)=\sum _{i\in [1,m],W(d_i,e_j)\ge \alpha ,W(d_i,e_k)\ge \alpha }\left({\displaystyle \frac{W\left(d_i,e_j\right)}{p{r}_j}}-{\displaystyle \frac{W\left(d_i,e_k\right)}{p{r}_k}}\right)\times {r_i}^{\prime }$$

(6)

where a higher absolute value of $C_c(p_j,p_k)$ indicates a greater difference in competitiveness between the two products.

3.4.2. Comparative Network Construction

To identify the competitive comparative relationships between products in the market, we constructed a directed weighted network called the PCAM-CPSS product’s competitive comparative network. Each product $p_i$ in the market is represented as a node, with the number of consumer reviews ($r_i$) serving as the node weight. The intensity of the comparative relationship $C_c(p_j,p_k)$ was used as the edge weight. If $C_c(p_j,p_k)\ge 0$, a directed edge is established from $p_j$ to $p_k$, indicating that $p_j$ is more competitive than or equal to $p_k$. Conversely, if $C_c\left(p_j,p_k\right)<0$, the edge is directed from $p_k$ to $p_j$, indicating that $p_j$ is less competitive than $p_k$.

3.4.3. Global Competitiveness Analysis

The PCAM-CPSS product’s competitive comparison network captures the comparative relationships among products in the target product’s market. To assess the overall competitiveness of products, we employed the weighted PageRank algorithm, a widely-used method for ranking nodes in directed weighted networks [12]. This algorithm evaluates the importance of nodes, allowing us to calculate the overall competitiveness of products in the market and rank their relative competitiveness:

$$PR\left(p_j\right)={\displaystyle \frac{1-\sigma }{m}}+\sigma \sum _{p_k\in O_{p_j}}\left({\displaystyle \frac{C_c\left(p_j,p_k\right)}{S_{in}\left(p_k\right)}}PR\left(p_k\right)\right)$$

(7)

where $PR\left(p_j\right)$ refers to the global competitiveness of the products in the competitive market; the higher value indicates stronger competitiveness. The damping coefficient is $\sigma (0<\sigma <1)$, and $m$ is the number of product nodes in the competitive market of the target product. $O_{p_j}$ refers to the node set that the nodes of product $p_j$ point to, and $S_{in}\left(p_k\right)$ refers to the input strength of the nodes of product $p_k$.

4. Model Evaluation

4.1. Case Selection and Data Acquisition

Smartphones, as widely used consumer electronics, possess well-defined and widely recognized attributes, with abundant user review data that are easy to collect and analyze. This makes them an ideal subject for research [57,58,59]. This research undertakes the validation of the PCAM-CPSS, employing JD (www.jd.com accessed on 1 September 2021) as the chosen e-commerce platform in China. The subject of analysis is the Huawei Mate 40 Pro smartphone, and the data collection took place on 1 September 2021. The dataset encompasses critical details such as the smartphone’s pricing and related information concerning other smartphones positioned within a specific price range. In the process, a substantial corpus of 21,000 reviews was meticulously analyzed, with 1000 reviews systematically harvested from the official flagship store for each respective smartphone. To ensure the robustness and comprehensiveness of our analysis, we conducted a comparative assessment of product attributes and consumer attention, drawing insights from both 1000 and 10,000 reviews for the Apple iPhone 12, sourced from multiple stores on JD. This methodological approach sought to affirm the consistency of our findings across diverse datasets. Importantly, our investigation confirmed that the initial crawl of 1000 online post-purchase reviews stand as an effective and reliable means for validating our analytical model.

4.2. Model Feasibility Verification

We validated the feasibility of the PCAM-CPSS using the crawled dataset. The specific steps proposed in Section 3.1, Section 3.2, Section 3.3 and Section 3.4 were followed for the validation process. To evaluate the effectiveness of the PCAM-CPSS, we compared it with two mainstream methods: attribute similarity-based ratings (AttrSim) [31] and co-occurrence statistics (Co-occur) [23].

4.2.1. Consumer Preference Satisfaction Measurement

The measurement of product attribute performance in three steps: Firstly, process online reviews using Stanford CoreNLP, extracting attribute words, modifiers, and opinion words to create tuples (product attribute, modifier, opinion word). Secondly, calculate sentiment scores, which results in a set of (product attribute word, sentiment score) pairs, using SnowNLP for opinion words, considering product attributes and modifiers. Thirdly, create an attribute dictionary for smartphones, incorporating information from the product manual.

Table 1. Attribute classification of the smartphones.

Attribute Category	Attribute Keywords
Appearance design	body, appearance, shape, sense of design, style…
Screen	screen, resolution, refresh rate, screen proportion…
Basic functions	signal, call, network, unlocking, transmission…
Camera functions	photography, video recording, camera, lens, zoom…
Entertainment functions	games, movies, videos, chasing dramas…
Memory	ROM, RAM, capacity, storage…
Accessories	accessories, mobile phone shell, earphone, data cable, protective film…
System	system, compatibility, bug, Android, IOS…
Performance	performance, operation, speed, fluency, CPU…
Audio	acoustics, sound quality, audio, speakers, bass…
Endurance and heating	endurance, battery, charging, power consumption, heating…
Services	service, after-sales, warranty, logistics, delivery…

presents the attribute classification.

The performance of various attributes was evaluated for each product based on attribute word classification. The results are presented in

Table 2. Examples of attribute performance and consumer attention across product categories.

Attribute Category	(A) Examples of the Performance of Various Attributes for Products			(B) Examples of Various Consumer Groups Paying Attention to Attributes
	$\mathbf{Apple} \mathbf{iPhone} 12 \left({\mathit{e}}_1\right)$	$\mathbf{Xiaomi} \mathbf{MIX}4 \left({\mathit{e}}_2\right)$	$\mathbf{Huawei} \mathbf{Mate} 40 \mathbf{Pro} \left({\mathit{e}}_3\right)$	$\mathbf{Attributes} 1 \left({\mathit{d}}_1\right)$	$\mathbf{Attributes} 2 \left({\mathit{d}}_2\right)$	$\mathbf{Attributes} 3 \left({\mathit{d}}_3\right)$
Appearance design	0.182249	0.111795	0.167376	0.322138	0.261811	0.203540
Screen	0.039505	0.041900	0.049677	0.087039	0.131398	0.116519
Basic functions	0.020555	0.003484	0.038960	0.041627	0.054626	0.047198
Camera functions	0.032887	0.023954	0.043272	0.070956	0.122539	0.026549
Entertainment functions	0.005041	0.002294	0.005178	0.011353	0.010335	0.081121
Memory	0.002217	0.002012	0.002656	0.009461	0.009350	0.017699
Accessories	0.001713	0.000925	0.003115	0.008988	0.008366	0.004425
System	0.024422	0.010423	0.027668	0.044465	0.041339	0.033923
Performance	0.108589	0.053971	0.106521	0.187323	0.137795	0.202065
Audio	0.047277	0.012630	0.035732	0.085620	0.066929	0.069322
Endurance and heating	0.020602	0.016142	0.047027	0.067171	0.106791	0.151917
Services	0.032922	0.035266	0.030525	0.063860	0.048720	0.045723

Note: After rounding the relevant floating-point numerical results, retention of six decimal places is maintained.

A. Huawei Mate 40 Pro excels in camera function, accessories, endurance, and heating compared to the other two smartphones. It is on par with Apple iPhone 12 across different attribute categories. However, Xiaomi Mix4 falls behind in all categories when compared to the other two products. The attention of consumer groups with similar needs on various product attributes was measured by calculating the frequency of attribute words extracted from online post-purchase reviews. The results are presented in Table 2B, revealing notable differences in consumer attention across the three categories. Consumers in the first category exhibit a higher focus on product appearance design compared to the other two categories. Consumers in the second category prioritize camera functions, while those in the third category emphasize entertainment features, endurance, and heating, particularly for seamless video playback and gaming experiences.

4.2.2. Competitive Relationship Network Construction

As the delineations in Section 3.2, we computed the competitive relationship intensity of the PCAM-CPSS method, specifically the variant incorporating price adjustment. This analytical endeavor commences with an assessment of consumer preference satisfaction, to unveil the congruence between products and the varied needs of heterogeneous consumer categories. Guided by Formula (3), the first quartile of consumer preference satisfaction, denoted as α, was duly established as the requisite threshold. This pivotal threshold enabled the subsequent computation of competitive relationship intensities among products. Consequently, the inter-product competitive relationship weights were meticulously recalibrated in accordance with the principles enshrined in Formula (5). This recalibration process is integral in accounting for the multifaceted influence of product pricing on consumer need. In the final phase, the assembly of the PCAM-CPSS product competitive relationship network was executed. A methodical approach was adopted, wherein the threshold β, representing the mean intensity value derived from the comprehensive spectrum of competitive relationships across all products, was meticulously applied. This judicious deployment of the β threshold effectively sieves out comparatively weaker relationships that might otherwise introduce unwarranted interference. Consequently, it culminates in the synthesis of a finely-honed product competitive relationship network of CPSS. In contrast, the PCAM-CPSS method without price adjustment is distinguished solely by the omission of price considerations in the calculation of inter-product competitive relationships. It parallels the foundational structure of the PCAM-CPSS product competitive intelligence analysis model.

Furthermore, the AttrSim product competitive analysis method engages in a two-step computational process. Initially, it conducts individual assessments of attribute performance for each product. Subsequently, it quantifies the intensity of competitive relationships among products by gauging the cosine similarity of attribute performance between each pair of products. Conversely, the Co-occur product competitive analysis method adopts a distinct approach. It quantifies the intensity of competitive relationships among products by drawing upon frequency statistics. These statistics are meticulously derived from the empirical examination of product co-occurrence within the corpus of online consumer reviews. This empirical foundation underpins the robust assessment of the strength of competitive relationships among products.

The outcomes of these four distinct methods, with respect to the intensity of product competitive relationships, are meticulously detailed in

Table 3. Competitors of the Huawei Mate 40 Pro and the corresponding competitive relationship intensity.

Product	PCAM-CPSS (With Price Adjustment)		PCAM-CPSS (Without Price Adjustment)		AttrSim		Co-Occur
	Intensity of Competition	Rank	Intensity of Competition	Rank	Intensity of Competition	Rank	Intensity of Competition	Rank
Apple iPhone 12	703.498642	1	15.64191	1	0.984386	5	74	1
Apple iPhone 12 mini	236.344165	6	4.620682	6	0.965799	14	\	\
Honor Magic3 Pro	218.283053	9	\	\	0.978498	9	19	3
Huawei Mate 40E	\	\	\	\	0.980313	8	\	\
Huawei P50 Pro	\	\	\	\	0.958249	17	37	2
Meizu 18 Pro	221.380203	8	\	\	\	\	\	\
Nubia 6Pro	\	\	\	\	0.970698	12	\	\
Nubia Z30Pro	\	\	\	\	0.995909	1	\	\
One Plus 9 Pro	514.333259	2	5.99635	2	0.973959	10	\	\
OPPO Find X3 Pro	314.987696	4	5.194045	4	0.982525	6	\	\
ROG 5s	\	\	\	\	\	\	\	\
Samsung Galaxy S20 Ultra	\	\	\	\	0.966945	13	4	5
Samsung Galaxy S21+	\	\	\	\	0.961575	16	4	6
Sony Xperia1 II	\	\	\	\	0.957284	18	\	\
vivo iQOO 8 Pro	\	\	\	\	0.982263	7	\	\
vivo X60t Pro+	209.803267	10	\	\	0.973361	11	\	\
Xiaomi 11 Pro	359.916932	3	5.0683	5	0.984832	3	6	4
Xiaomi 11 Ultra	310.221226	5	5.917225	3	0.984443	4	3	7
Xiaomi MIX4	228.586855	7	\	\	0.965501	15	\	\
ZTE Axon 30Ultra	\	\	\	\	0.985566	2	\	\

Note: After rounding the relevant floating-point numerical results, retention of six decimal places is maintained.

.

4.2.3. Competitive Market Segmentation

The Competitive Market Segmentation Network for the four aforementioned methods is presented in Figure 3. The utilization of the SLPA algorithm in delineating competitive markets within the PCAM-CPSS product’s competitive relationship network is marked by its convergence after a sequence of 20 iterations. An integral component of this process is the filter threshold associated with node labels, which significantly influences the demarcation of products spanning multiple communities. A higher threshold signifies the presence of more robust relationships, while a lower threshold paves the way for a finer-grained market segmentation. In our empirical context, a filter threshold of 0.1 was thoughtfully chosen to align with established values. Network visualization was seamlessly executed through the Graph-tool, casting consumer reviews as nodes and competitive relationships as edges. Product nodes share a consistent color scheme, with products belonging to multiple communities portrayed in distinct shades to signify instances of overlapping communities. In contrast, the PCAM-CPSS without price adjustment stands out solely due to its omission of price, maintaining uniformity with other considerations.

Furthermore, within the domain of the AttrSim product competitive analysis method, an analogous initial screening process unfolds, driven by pricing considerations. Selected products serve as nodes, with node weights contingent on the volume of consumer evaluations. Subsequent steps involve the creation of undirected weighted edges between products, with edge weights intricately tied to the strength of competitive relationships, thereby forming a network that encapsulates product attribute performance similarity. Meanwhile, the Co-occur product competitive analysis method follows a parallel trajectory, initiating with price-screened products as network nodes. Node weights reflect consumer evaluations, and directed weighted edges emerge based on references to competing products within product reviews, signifying the intensity of competitive relationships and shaping a product co-occurrence network. In both methodologies, edges characterized by weaker competitive relationship intensities are systematically filtered, with the mean edge weight serving as the criterion for exclusion. Concurrently, the SLPA overlapping community discovery algorithm facilitates market segmentation in both instances.

4.2.4. Global Product Competitiveness Analysis

As depicted in Figure 3a, the competitive market delineates a cluster comprising Apple iPhone 12, Huawei Mate 40 Pro, One Plus 9 Pro, Xiaomi 11 Ultra, Xiaomi 11 Pro, and OPPO Find X3 Pro, manifesting notably proximate competitive interrelations. This intricate web of competitive relationships necessitates a meticulous quantitative assessment, employing Formula (6) to gauge the strength of comparative competitiveness among these products. This assessment takes into account the weighted disparities that underscore the products’ varying degrees of alignment with the manifold needs of diverse consumer segments. Subsequently, underpinned by the discerned comparative competitiveness relationships and their respective magnitudes, we constructed the PCAM-CPSS product competitiveness comparison network. In the denouement, we invoked the weighted PageRank algorithm to derive a comprehensive evaluation of each product’s global competitiveness within this competitive market, leading to the ordered ranking. In contrast, the PCAM-CPSS method without price adjustment stands out solely due to its omission of price-related factors. The comparative competitive network generated by both methods is depicted in Figure 4. The comprehensive evaluation of products within the competitive market encompassing Huawei Mate 40 Pro is detailed in

Table 4. Overall competitiveness in the competitive market of Huawei Mate 40 Pro.

Product	PCAM-CPSS (With Price Adjustment)		PCAM-CPSS (Without Price Adjustment)		Co-Occur
	Intensity of Competition	Rank	Intensity of Competition	Rank	Competitive Relationship Intensity	Rank
Apple iPhone 12	0.405819	1	0.249550	1	0.258509	1
Apple iPhone 12 mini	\	\	0.045197	7	0.007143	21
Honor Magic3 Pro	\	\	0.033373	11	0.061255	5
Huawei Mate 40 Pro	0.204913	2	0.139731	2	0.180074	2
Huawei Mate 40E	\	\	0.028961	12	0.127380	4
Huawei P50 Pro	\	\	0.028367	13	0.176769	3
Meizu 18 Pro	\	\	0.037521	8	0.007520	19
Nubia 6Pro	\	\	0.022514	15	0.007690	16
Nubia Z30Pro	\	\	0.081887	3	0.007854	15
One Plus 9 Pro	0.128376	3	0.059769	5	0.008007	14
OPPO Find X3 Pro	0.083310	5	\	\	0.008924	11
ROG 5s	\	\	\	\	0.007688	17
Samsung Galaxy S20 Ultra	\	\	0.021301	17	0.016424	9
Samsung Galaxy S21+	\	\	\	\	0.027582	7
Sony Xperia1 II	\	\	\	\	0.007143	20
vivo iQOO 8 Pro	\	\	0.021784	16	0.007591	18
vivo X60t Pro+	\	\	0.034611	10	0.008537	12
Xiaomi 11 Pro	0.095091	4	0.055417	6	0.028473	6
Xiaomi 11 Ultra	0.082491	6	0.078698	4	0.024256	8
Xiaomi MIX4	\	\	0.036454	9	0.013140	10
ZTE Axon 30Ultra	\	\	0.024865	14	0.008042	13

Note: After rounding the relevant floating-point numerical results, retention of six decimal places is maintained.

.

However, the AttrSim product competitive analysis method, which relies on product attribute performance similarity, lacks the capacity to provide a comprehensive assessment of each product’s overall competitiveness. Consequently, it is not a suitable benchmark for evaluating the accuracy of overall product competitiveness. In contrast, the Co-occur method, based on product co-occurrence in online reviews, offers partial insights into a product’s competitive standing. This method evaluates a product’s global competitiveness by considering its centrality within the product co-occurrence network, mirroring the approach of the PCAM-CPSS method. This study utilizes the weighted PageRank algorithm to calculate and rank the global competitiveness of each product, as documented in Table 4.

4.3. Model Effectiveness Evaluation

This research undertakes an evaluation of the efficacy of the PCAM-CPSS, in comparison with two alternative competitive analysis methods. The primary objective is to gauge the precision of these methodologies in the context of competitor identification and the quantification of product competitiveness. In pursuit of this objective, an enhanced NDCG index was employed to facilitate a comparative assessment of the accuracy in the identification of competitor products and the ranking of overall product competitiveness derived from diverse methodologies.

Drawing on prior research regarding the identification of product competitors [60], the current study adopts a querying approach denoted as “Huawei Mate 40 Pro and product $i$” within the Baidu website. It employs the count of web pages retrieved through querying specific keywords on a search engine as a benchmark, assessing the strength of the competitive association between Huawei Mate 40 Pro and alternative products. Furthermore, an evaluation of the precision in ranking the competitive intensity between products, as accomplished through various methods, was performed. This evaluation leveraged the enhanced NDCG index to compare the efficacy of distinct approaches for competitor identification. Moreover, the NDCG index compared the obtained ranking results with the benchmark ranking result by calculating the cumulative income, which has been widely used in recommendation systems, information retrieval, and competitiveness analysis [60,61]. However, the NDCG index only takes 0 or 1 as the result when calculating the correlation between the ranking results and the benchmark ranking and does not further accurately quantify the difference between them. Hence, accurately measuring the difference between the ranking results obtained by different methods and the benchmark ranking result, the NDCG index was improved by further refining $rel\left(c\right)$:

$$N{DCG}_e={\displaystyle \frac{1}{Z}}\sum _{c\in C}{\displaystyle \frac{2^{rel\left(c\right)}-1}{{log}_2\left(1+BInd\left(c\right)\right)}}$$

(8)

$$rel\left(c\right)={\displaystyle \frac{1}{{\mathit{log}}_2(2+\left|Ind\left(c\right)-BInd\left(c\right)\right|)}}$$

(9)

$$Z=\sum _{c\in C}{\displaystyle \frac{1}{{log}_2\left(1+BInd\left(c\right)\right)}}$$

(10)

where $e$ represents the target product and $C$ represents the identified competitive product set. $Ind\left(c\right)$ indicates the competitive relationship intensity ranking of competitive product $c$ within the competitive product set $C$ for the target product $e$. $BInd\left(c\right)$ represents the benchmark ranking of competitive product c within the competitive product set $C$ for the target product $e$. The value of $rel\left(c\right)$ reflects the correlation between the competitive relationship intensity ranking and the benchmark ranking of competitive product $c$. A higher NDCG index indicates greater accuracy in ranking results, indicating the effectiveness of the competitive analysis method in identifying competitors.

The outcomes of the $NDCG$ evaluations concerning the ranking of competitive relationship intensity for each method are showcased in

Table 5. $NDCG$ results of competitive relationship intensity ranking obtained via each method.

Abbreviation of the Method	$\mathit{N}\mathit{D}\mathit{C}\mathit{G}$
AttrSim	0.400286
Co-occur	0.617927
PCAM-CPSS (without price adjustment)	0.580849
PCAM-CPSS (with price adjustment)	0.637584

Note: After rounding the relevant floating-point numerical results, retention of six decimal places is maintained.

. Notably, the PCAM-CPSS exhibits the highest level of precision in competitor identification, with the price-adjusted variant securing the second position. The Co-occur method emerges as the second most accurate in this regard. However, it is essential to underscore that the AttrSim method lags significantly in terms of accuracy, registering the lowest levels among the three methodologies, notably trailing behind the others.

The ranking of product sales within the identified competitive market serves as the reference standard for competitiveness assessment. To gauge the efficacy of distinct methodologies in competitiveness analysis, the enhanced $NDCG$ index was deployed. It is imperative to underscore that the AttrSim approach, hinging on attribute similarity for competitiveness evaluation, was excluded from comparative analysis due to its inherent limitation in reflecting individual product competitiveness. Within the scope of this evaluation, the PCAM-CPSS attains the highest degree of precision in both extensive and limited communities, with notably superior accuracy discerned in the context of smaller communities, as depicted in

Table 6. $NDCG$ results of product’s competitiveness ranking obtained via each method.

Abbreviation of the Method	$\mathit{N}\mathit{D}\mathit{C}\mathit{G}$
Co-occur	0.471047
PCAM-CPSS (without price adjustment)	0.607527
PCAM-CPSS (with price adjustment, large community)	0.629568
PCAM-CPSS (with price adjustment, small community)	0.888321

Note: After rounding the relevant floating-point numerical results, retention of six decimal places is maintained.

. However, it is noteworthy that the analysis of competitiveness, when decoupled from the influence of product price, registers a marginally diminished level of accuracy in comparison to the PCAM-CPSS model. Conversely, the Co-occur methodology manifests the lowest degree of accuracy in the domain of competitiveness analysis, markedly trailing behind alternative approaches.

5. Research Results

In this study, we introduced the innovative PCAM-CPSS model, which is based on the analysis of consumer preference satisfaction similarity derived from post-purchase online reviews. To assess the practical utility of the model, we selected some smartphones as our primary case study. This intentional choice allows for a focused evaluation of the model’s feasibility and performance. The results confirm the effectiveness of the PCAM-CPSS in product competitive analysis, with key findings including the following:

(1)

The model confirms that consumer preference satisfaction plays a critical role in product assessment and selection within specific price constraints. By quantifying this satisfaction, the PCAM-CPSS facilitates the discernment of competitive relationships, market boundaries, and product competitiveness.

(2)

Compared to existing models such as AttrSim and Co-occur, the PCAM-CPSS shows superior performance in identifying competitors and evaluating product competitiveness. The model addresses the limitations of attribute similarity and product co-occurrence, enhancing the accuracy of competitive analysis.

(3)

The results underscore the importance of incorporating price as a factor in product analysis. Ignoring price can lead to inaccurate evaluations of consumer satisfaction, particularly for products that are competitively priced. Including price dynamics enhances the reliability of competitiveness assessments.

6. Conclusions and Implications

The introduction of the PCAM-CPSS model offers valuable insights into product competitive analysis, particularly in the context of UGC. The findings from this study suggest several key implications for both research and practice:

(1)

This study confirms that consumer preference satisfaction is a critical determinant in product assessment, particularly when considered within specific price ranges. By quantifying satisfaction, the PCAM-CPSS model helps identify competitive relationships, define market boundaries, and assess product competitiveness more accurately. This highlights the importance of integrating consumer satisfaction into competitive analysis models, encouraging researchers to explore similar approaches across different industries and product categories.

(2)

The model’s superior performance compared to existing approaches like AttrSim and Co-occur suggests that a more comprehensive framework that incorporates both product attributes and co-occurrence can lead to more accurate competitor identification and competitiveness evaluation. Future research could focus on further refining the model to capture even more granular details of competitor relationships, potentially applying the model to more complex product ecosystems or markets.

(3)

The model’s ability to identify key competitive attributes and align them with consumer satisfaction provides a valuable tool for businesses looking to optimize their product offerings. Companies can leverage these insights to prioritize features and attributes that resonate most with consumers, ultimately positioning themselves for long-term market success and sustained competitive advantage.

7. Limitations and Future Work

While the PCAM-CPSS model offers significant improvements in competitor identification and competitiveness evaluation within the UGC framework, several limitations should be considered, particularly related to UGC data quality, sentiment analysis, and the model’s scope of application.

(1)

UGC data can be inconsistent and subjective, with conflicting opinions or unclear language that may reduce analysis accuracy. For example, reviews may include conflicting opinions, ambiguous language, or lack sufficient detail, making it difficult for the model to extract reliable product attribute terms. To improve the model, future research could focus on refining data cleaning, filtering, and validation methods to enhance data quality and robustness.

(2)

Sentiment analysis in the model faces difficulties with nuances such as irony, sarcasm, and culturally specific expressions. Misinterpretation of such nuances may lead to inaccurate sentiment classification. For instance, a review that uses ironic language could easily be misclassified as negative or positive, even if the sentiment expressed does not align with the content. Future improvements could involve more advanced NLP techniques to better handle these complexities.

(3)

The current manual classification of product attributes, while ensuring accuracy, is time-consuming and limits scalability. The reliance on human processing for attribute classification can introduce inconsistencies, as different annotators may interpret attributes in slightly varying ways. Additionally, manual intervention can become a bottleneck when the model is applied to large datasets. Automating the classification process could improve efficiency and scalability while maintaining accuracy.

(4)

While the evaluation of a single product provides valuable insights into the feasibility of the model, it may not fully capture the complexities and variations that arise when the model is applied to a diverse set of products or across different industries. Expanding the model’s application to a broader range of products and industries in future studies would help test its adaptability, reliability, and scalability, providing further validation and identifying potential domain-specific improvements.