Product strategy improvement method based on LSGDM-SHAP

Abstract

The invention provides a product strategy improvement method and system based on LSGDM-SHAP, a storage medium and electronic equipment, and relates to the technical field of natural language processing. The method comprises the following steps: firstly, collecting online comparison comments published by a plurality of consumers after purchasing a target product or a competitive product thereof; secondly, attribute information in the online comparison comments is extracted, and consumer sentiment analysis is carried out; thirdly, an LSGDM method is adopted to seek a consensus among different consumer viewpoints so as to obtain product attribute performance expression which can best represent the holistic viewpoint; adopting an SHAP interpretability method to obtain the importance of product attributes; and finally, mapping product attributes to a two-dimensional plane under an IPA model framework to obtain a product attribute improvement strategy of the target product. From the perspective of market competition, on-line comparison comments are utilized to quickly and accurately understand consumer demands, help enterprises to grasp product promotion key points and provide product promotion strategies for the enterprises, so that enterprise resources are effectively and reasonably utilized.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and particularly relates to a product strategy improvement method, system, storage medium, and electronic device based on LSGDM-SHAP. Background Art

[0002] With the diversified and personalized development of consumer demands, the product iteration frequency has accelerated, and the development cycle has been shortened. Whether an enterprise can quickly understand user demand preferences and accurately grasp the product improvement direction has become a key link in the product development process.

[0003] In the prior art, the IPA analysis model is a commonly used business research technology for understanding customer needs and formulating product improvement strategies. The traditional IPA model divides product attributes from the consumer perspective into four quadrants using two dimensions of importance and performance through market research, helping enterprises grasp the product improvement direction. However, the traditional IPA model usually collects data from consumers by means of questionnaires or on-site visits, which is time-consuming and laborious and cannot adapt to the current fierce market competition environment.

[0004] For example, existing researchers used hotel online reviews and ratings as data to conduct IPA model analysis on hotel services to help hotels clarify service type improvement strategies. The research first extracts hotel service attributes mentioned by consumers from online reviews and conducts sentiment analysis, taking the sentiment result of consumers towards hotel services as the performance of the hotel service. Then, the research constructs an integrated neural network model using the sentiment results and scoring of consumers to measure the importance of hotel services. Finally, an IPA model is constructed based on the measured performance and importance, dividing the service attributes mentioned by consumers into four quadrants and providing strategies for hotels to improve services.

[0005] However, the original product improvement strategy model only analyzes a single product or service and gives product improvement suggestions. Summary of the Invention

[0006] (1) Technical Problems to be Solved

[0007] Aiming at the deficiencies of the prior art, the present invention provides a product strategy improvement method, system, storage medium, and electronic device based on LSGDM-SHAP, which solves the technical problem that the original product improvement strategy model only analyzes a single product or service.

[0008] (2) Technical Solutions

[0009] To achieve the above objectives, the present invention is realized through the following technical solutions:

[0010] A product strategy improvement method based on LSGDM-SHAP, comprising:

[0011] Collecting a number of online comparison reviews published after consumers purchase the target product or its competing products; each of the online comparison reviews includes attribute comparison information of the target product and the competing products;

[0012] Extracting the attribute information from the online comparison reviews, and performing consumer sentiment analysis on the attribute information to obtain the sentiment scores of different product attributes in each of the online comparison reviews;

[0013] Based on the sentiment scores, using the LSGDM method to seek consensus among different consumer viewpoints to obtain the product attribute performance that best represents the overall viewpoint;

[0014] Based on the sentiment scores and the purchase results, pre-training a neural network model corresponding to each product attribute, using the SHAP interpretability method to calculate the Shapley value of each product attribute, and multiplying it by the sentiment score to obtain the contribution value of different product attributes to the predicted purchase result of the model, and taking it as the product attribute importance;

[0015] Based on the product attribute performance and the product attribute importance, mapping the product attributes to a two-dimensional plane under the IPA model framework to obtain the product attribute improvement strategy of the target product.

[0016] Preferably, LDA is used to extract the attribute information from the online comparison reviews.

[0017] Preferably, the performing consumer sentiment analysis on the attribute information includes:

[0018] Using the conditional random field method to identify the product names;

[0019] Using the Snownlp tool to perform sentiment analysis, and representing it with a number between 0 and 1. If the score is greater than 0.5, it means the sentiment is positive; otherwise, if the score is less than 0.5, it means the sentiment is negative.

[0020] Preferably, the based on the sentiment scores, using the LSGDM method to seek consensus among different consumer viewpoints to obtain the product attribute performance that best represents the overall viewpoint includes:

[0021] Using the Kmeans algorithm to cluster all the online comparison reviews into multiple sub-group viewpoints, and calculating the comprehensive expression of each sub-group viewpoint;

[0022] By constructing a minimum variance objective function, continuously optimizing the opinion weights of each sub-group viewpoint;

[0023] Based on the best opinion weights of each of the subgroup views and the adjusted comprehensive expression, a weighted evaluation reaches an overall decision consensus, which is used as the measurement result of the product attribute performance that best represents the overall view.

[0024] Preferably, the SHAP interpretability method is used to calculate the Shapley value of each product attribute, and the calculation formula is:

[0025]

[0026] Where, is the Shapley value of attribute A i , F is the set of all attribute sentiment scores; S is the subset of attributes that does not include A i , f(S) is the predicted value of the model on the attribute subset S, and f(S ∪ {A i}) - f(S) is the marginal contribution of attribute A i to the subset S; is the weight, which is used to ensure that the contributions of all product attribute combinations are fairly distributed.

[0027] Preferably, the product attribute improvement strategy includes:

[0028] Respectively set the division thresholds for the importance of product attributes and the performance of product attributes;

[0029] If both the importance of the product attribute and the performance of the product attribute are higher than the corresponding thresholds, the product attribute improvement strategy is to continue to maintain;

[0030] If the importance of the product attribute is higher than the corresponding threshold and the performance of the product attribute is lower than the corresponding threshold, the product attribute improvement strategy is to focus on;

[0031] If the importance of the product attribute is lower than the corresponding threshold and the performance of the product attribute is higher than the corresponding threshold, the product attribute improvement strategy is over - effort;

[0032] If both the importance of the product attribute and the performance of the product attribute are lower than the corresponding thresholds, the product attribute improvement strategy is low priority.

[0033] A product strategy improvement system based on LSGDM - SHAP includes:

[0034] A comment collection module, which is used to collect a number of online comparison comments published by consumers after purchasing the target product or its single competing product; each of the online comparison comments includes the attribute comparison information between the target product and the single competing product;

[0035] An emotion analysis module, configured to extract attribute information from the online comparison reviews, and perform consumer emotion analysis on the attribute information to obtain emotion scores of different product attributes under each purchase result;

[0036] A performance acquisition module, configured to seek consensus among different consumer opinions by using the LSGDM method based on the emotion scores, so as to obtain the product attribute performance that best represents the overall opinion;

[0037] An attribute importance acquisition module, configured to pre-train a neural network model corresponding to each product attribute based on the emotion scores and purchase results, calculate the Shapley value of each product attribute by using the SHAP interpretability method, and multiply it by the emotion scores to obtain the contribution value of different product attributes to the predicted purchase result of the model, and use it as the product attribute importance;

[0038] An improvement strategy acquisition module, configured to map product attributes to a two-dimensional plane under the IPA model framework based on the product attribute performance and the product attribute importance, so as to obtain the product attribute improvement strategy of the target product.

[0039] A storage medium stores a computer program for improving product strategies based on LSGDM-SHAP, wherein the computer program enables a computer to execute the product strategy improvement method as described above.

[0040] An electronic device includes:

[0041] One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the programs include those for executing the product strategy improvement method as described above.

[0042] (III) Advantageous Effects

[0043] The present invention provides a product strategy improvement method, system, storage medium and electronic device based on LSGDM-SHAP. Compared with the prior art, the following advantageous effects are achieved:

[0044] In the present invention, first, a number of online comparison reviews published by consumers after purchasing the target product or its competing products are collected; second, the attribute information in the online comparison reviews is extracted and consumer sentiment analysis is performed; third, the LSGDM method is used to seek consensus among different consumer views to obtain the product attribute performance that best represents the overall view; then, the SHAP interpretability method is used to obtain the importance of product attributes; finally, the product attributes are mapped to a two-dimensional plane under the IPA model framework to obtain the product attribute improvement strategy for the target product. From the perspective of market competition, by using online comparison reviews, the consumer needs can be quickly and accurately understood, which helps enterprises grasp the key points of product improvement, provides product improvement strategies for enterprises, and enables the effective and reasonable utilization of enterprise resources. Brief Description of the Drawings

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following-described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0046] Figure 1 It is a block diagram of a product strategy improvement method based on LSGDM-SHAP provided by an embodiment of the present invention;

[0047] Figure 2 It is a flowchart of a product strategy improvement method based on LSGDM-SHAP provided by an embodiment of the present invention;

[0048] Figure 3 It is an example diagram of an IPA model provided by an embodiment of the present invention. Detailed Embodiments

[0049] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0050] By providing a product strategy improvement method, system, storage medium, and electronic device based on LSGDM-SHAP, the embodiments of the present application solve the technical problem that the original product improvement strategy model only analyzes a single product or service.

[0051] The key points of the technical solutions in the embodiments of the present application to solve the above technical problems are as follows:

[0052] 1. A product improvement strategy considering the competitive environment is provided. In the embodiments of the present invention, reviews are classified into two types: general reviews and comparative reviews. Comparative reviews are used as research data. By comparing two products by consumers and the final purchase results, the competitive advantages and disadvantages between products are analyzed, and product improvement strategies are provided.

[0053] 2. The consensus of consumers is obtained using the large-scale group decision-making method. Through the idea of large-scale group decision-making, consensus is sought among different consumer opinions, and the measurement results of the product attribute performance that best represent the overall view are obtained.

[0054] 3. An interpretable neural network model is used to measure the importance of product attributes. Through the SHAP interpretability technique, the contribution value of each product attribute to the final purchase result is calculated, and then the importance of product attributes is measured, solving the problem of poor interpretability of traditional neural networks.

[0055] To better understand the above technical solutions, the above technical solutions will be described in detail below in conjunction with the accompanying drawings of the specification and specific implementation manners.

[0056] Example 1:

[0057] As Figure 1 shown, the embodiments of the present invention provide a product strategy improvement method based on LSGDM-SHAP, including:

[0058] S1. Collect online comparative reviews published by several consumers after purchasing the target product or its competing product; each of the online comparative reviews includes attribute comparison information of the target product and the competing product;

[0059] S2. Extract the attribute information in the online comparative reviews, and perform consumer sentiment analysis on the attribute information to obtain the sentiment scores of different product attributes in each of the online comparative reviews;

[0060] S3. Based on the sentiment scores, use the LSGDM method to seek consensus among different consumer opinions to obtain the product attribute performance that best represents the overall view;

[0061] S4. Based on the sentiment scores and purchase results, pre-train a neural network model corresponding to each product attribute, use the SHAP interpretability method to calculate the Shapley value of each product attribute, and multiply it by the sentiment score to obtain the contribution value of different product attributes to the predicted purchase result of the model, and use it as the importance of product attributes;

[0062] S5. Based on the product attribute performance and the importance of product attributes, map the product attributes to a two-dimensional plane under the IPA model framework to obtain the product attribute improvement strategy of the target product.

[0063] From the perspective of market competition, the embodiments of the present invention utilize online comparative reviews to quickly and accurately understand consumer needs, help enterprises grasp the key points of product improvement, provide product improvement strategies to enterprises, and enable the effective and reasonable utilization of enterprise resources.

[0064] As Figure 2 shown, Figure 2 a flowchart for improving product strategy based on LSGDM-SHAP is disclosed.

[0065] Next, each step of the above solution will be introduced in detail in combination with Figure 2 :

[0066] With the activity of network platforms, consumers express their product usage experiences through online reviews on social media software. Due to the randomness of consumers' language, there is a lot of useless information in online reviews, such as modal particles, emojis, and expressions unrelated to the product.

[0067] In addition, unstructured text language is difficult to directly apply to subsequent analysis. Therefore, it is necessary to extract key information for analysis from online reviews and convert it into structured data.

[0068] Specifically, refer to the detailed introduction in steps S1 to S3:

[0069] In step S1, collect a number of online comparative reviews published by consumers after purchasing the target product or its competing products; each of the online comparative reviews includes attribute comparison information between the target product and the competing product.

[0070] It should be noted that the embodiments of the present invention classify online reviews into two types: general reviews and comparative reviews. This step uses online comparative reviews as research data. Online comparative reviews refer to that when consumers publish online reviews, in addition to expressing their experience of the product, they often compare the product with the competing products in their minds in multiple aspects. Such comparative reviews can reflect the influence degree of different attributes of the two products on consumers' purchase decisions.

[0071] Exemplarily, since the market environment of the automotive product is highly competitive and has rich functional attributes, the embodiments of the present invention take automotive products as an example, select 2 best-selling compact automotive products, and collect the word-of-mouth data of the two automotive products from three websites: Autohome, Dongchedi, and Pacific Auto.

[0072] In step S2, extract the attribute information from the online comparative reviews, and perform consumer sentiment analysis on the attribute information to obtain the sentiment scores of different product attributes in each of the online comparative reviews.

[0073] In an alternative embodiment, LDA is used to extract the attribute information from the online comparison reviews. LDA is an unsupervised machine learning method based on Bayesian probability, which is used to represent multiple topics in a document in the form of a probability distribution. Research shows that LDA can effectively extract the attributes of products from online reviews.

[0074] To improve the extraction effect of product attributes, some preprocessing is performed on the online reviews in this step. Usually, product attributes usually appear in user reviews in the form of nouns or noun phrases. Therefore, using the method of part-of-speech tagging, the nouns and noun phrases in the online reviews are retained, and adjectives, adverbs, punctuation marks, etc. are removed.

[0075] In particular, there are some hierarchical relationships in automotive attributes. For example, the "space" attribute includes fine-grained attribute expressions such as "front and rear rows", "cab space", and "legs". Therefore, considering the professionalism of the automotive field, in an alternative embodiment, this step also manually constructs an attribute hierarchy dictionary with the help of domain expert knowledge to further adjust the results extracted by LDA.

[0076] Furthermore, in an alternative embodiment, the consumer sentiment analysis for the attribute information includes:

[0077] Using the conditional random field method to identify the product name;

[0078] Using the Snownlp tool for sentiment analysis and representing it with a number between 0 and 1. If the score is greater than 0.5, it means the sentiment is positive; otherwise, if the score is less than 0.5, it means the sentiment is negative.

[0079] Specifically, in product comparison reviews, consumers will express their opinions on the performance of products in certain attributes and compare them with other products. Therefore, at least two or more product names will be mentioned in the comparison reviews. In addition, considering the randomness of consumers' language, there will also be review types that only mention the comparison product names. Therefore, the comparison reviews studied in the embodiments of the present invention refer to consumers' online reviews that mention other products in addition to the purchased product itself. First, we use the conditional random field method to identify the product name. Considering that consumers often use a large number of product nicknames and Internet terms to refer to products, in an alternative embodiment, a product name dictionary is also manually constructed to more accurately identify whether and which product names are mentioned in the product reviews.

[0080] In comparative reviews, consumers express their emotions towards product attributes on the premise of comparing with other products. Exemplarily, taking two products A and B as an example, if product B is mentioned in the review of consumer's purchase of product A, it is considered that the consumer compares the two products. On this basis, if the consumer's emotion towards a certain attribute of product A is positive, it is considered that the consumer thinks product A is superior to product B in this attribute, denoted as 1; if the consumer expresses negative emotion towards a certain attribute, then the consumer will think product A is inferior to product B in this attribute, denoted as -1; if no attribute is mentioned, it is denoted as 0.

[0081] On this basis, in this step, the Snownlp tool is used for sentiment analysis. Snownlp can calculate the sentiment score of a sentence and represent it with a number between 0 and 1. If the score is greater than 0.5, it means the sentiment is positive; on the contrary, if the score is less than 0.5, it means the sentiment is negative. Finally, the performance of product attributes that can best represent the overall view as shown in Table 1 can be obtained.

[0082] Table 1

[0083] Comment Number Purchase Result Attribute 1 Attribute 2 … Attribute n 1 A 1 -1 … 0 2 B 0 -1 … -1 … … … … … … n A 1 -1 … 0

[0084] In step S3, based on the sentiment score, the LSGDM method is used to seek consensus among different consumer views to obtain the performance of product attributes that can best represent the overall view.

[0085] The emotion of product attributes expressed in online reviews reflects the product performance perceived by consumers. Online reviews reflect the decision-making process of consumers. The emotion expression of product attributes and purchase results by different consumers in the reviews can be understood as the decision-making results made by different decision-makers after considering their respective preferred decision-making schemes. Large-scale group decision-making is a consensus reaching process that can obtain a consensus view from online reviews with different views.

[0086] Therefore, in this step, the large-scale group decision-making (LSGDM, Large-scale Group DecisionMaking) technology is used to measure the performance of product attributes that can accurately express the views of group consumers from online reviews. Generally speaking:

[0087] First, the Kmeans algorithm is used to cluster all online comparative reviews into multiple sub-group views and calculate the comprehensive expression of each sub-group view.

[0088] Second, by constructing a minimum variance objective function, the opinion weights of each sub-group view are continuously optimized.

[0089] Finally, based on the best opinion weights of each of the subgroup views and the adjusted comprehensive expression, a weighted evaluation is performed to reach an overall decision consensus, which is used as the measurement result of the product attribute performance that best represents the overall view.

[0090] Specifically:

[0091] 1) Decision subgroup division

[0092] For the convenience of expression, assume that there are a total of n online comparison comments r k (k = 1,..., n), and m product attributes A i (i = 1,..., m) are extracted from the online comparison comments. According to the sentiment scores perf i of each attribute in the online comments, the online comments can be expressed in vector form.

[0093] Calculate the consensus degree CD xy between two comments, and the calculation formula is as follows:

[0094]

[0095] Among them, are the sentiment scores of the product attribute A x of the online comparison comments r y and r i respectively.

[0096] Use the Kmeans algorithm to cluster the online comments, and use the consensus degree as the distance between the clustering objects. The entire group LG can be divided into w decision subgroups SG k (k = 1,..., w), that is, w subgroup views.

[0097] After clustering, it is necessary to calculate the comprehensive expression of each subgroup view. The comprehensive view of the subgroup can also be expressed in vector form. The calculation formula is as follows:

[0098]

[0099] Among them, n k is the number of comments in SG k , and r a is a comment belonging to SG k .

[0100] 2) Reaching a consensus

[0101] After obtaining the comprehensive expressions of the subgroup views, first calculate the internal cohesion of the subgroup, and the calculation formula is as follows:

[0102]

[0103] The cohesion within the subgroup and the comprehensive expression of the subgroup are evaluated with weights to obtain a preliminary overall decision-making plan. Among them,

[0104]

[0105] At this time, the overall decision-making plan has not reached a consensus. To achieve a consensus, it is necessary to identify one or several subgroups among the subgroups that have the greatest difference in opinions from other subgroups. These subgroups hinder the entire group from reaching a consensus and need to be adjusted.

[0106] Therefore, it is necessary to calculate the opinion consistency among the subgroups, which needs to be divided into attribute consistency and overall consistency. The calculation formulas are as follows:

[0107]

[0108] If is closer to 1, it indicates that the opinion of subgroup SG k is closer to the group. At this time, a consistency threshold α∈[0.5, 1) needs to be set to find those subgroups that need to be adjusted and have a large difference in opinions from the group.

[0109] Next, for the subgroups with consistency lower than α, adjustments are made. It is equivalent to that in the Delphi method, the subgroups with inconsistent opinions gradually unify their views through repeated exchanges. Assume that SG (k) is the subgroup with consistency lower than α, and the adjustment method is as follows:

[0110]

[0111] Among them, is the emotional expression lower than α on attribute A (k) in SG m , and δ r is the adjustment parameter used to control the intensity of adjusting the inconsistent subgroup SG (k) to be similar to the group opinion, which is provided by the decision maker in advance.

[0112] To achieve group consensus, it is necessary to add weights to the subgroup according to the cohesion within the subgroup, and finally obtain the group consensus through weighting. As the subgroup SG (k) is adjusted, the cohesion W(SG (k) ) of this subgroup will also change. Therefore, it is necessary to construct a minimum variance objective function to obtain the optimal weights W = [W(SG1), W(SG2),..., W(SG w )], and the objective function is calculated as follows:

[0113]

[0114]

[0115] By optimizing the model, the optimal weights of each subgroup can be obtained, and the weights are weighted with the comprehensive expressions of the adjusted subgroups to obtain the final group consensus expression. This group consensus can be regarded as the common view of all online reviews on the sentiment scores of various product attributes, that is, the performance of the product attributes considered by consumers to be measured.

[0116] In step S4, based on the sentiment score and the purchase result, a neural network model corresponding to each product attribute is pre-trained, the Shapley value of each product attribute is calculated by using the SHAP interpretability method, and it is multiplied by the sentiment score to obtain the contribution value of different product attributes to the predicted purchase result of the model, and it is used as the product attribute importance.

[0117] The importance of product attributes is reflected in the degree of influence of the sentiment comparison results of consumers on various product attributes on the final purchase decision, and there is a non-linear relationship between the two. Although traditional artificial neural networks can achieve very good results in fitting this non-linear relationship, the black-box nature of neural networks makes it impossible to clearly identify what factors affect the final model output. In order to clarify the parameter relationship inside the neural network, in an optional implementation manner, the embodiments of the present invention use the SHAP (SHapley Additive exPlanations) interpretability technology to obtain the importance degree of the influence of product attributes on the final purchase decision.

[0118] Specifically, this step includes:

[0119] 1) Construct a neural network model

[0120] Take the sentiment scores of consumers on different attributes as the input of the neural network, and take the purchase result of the user as the output, and train the neural network model. Since consumers do not mention all attributes in the comments, the product comparison result data is relatively sparse. Sparse data will affect the training result of the network. Therefore, using the idea of K-fold cross-validation, the data set is divided into K parts and used to train K neural network models to avoid the overfitting problem of the model caused by sparse data.

[0121] 2) Calculate the importance of product attributes

[0122] SHAP is an interpretation method based on the Shapley value. It explains the prediction results of the model by calculating the Shapley value of each feature. The core idea of the Shapley value is to calculate the marginal contribution of each feature in all possible feature combinations and perform a weighted average on it. For attribute A i , its Shapley value calculation formula is:

[0123]

[0124] where, is the Shapley value of attribute A i , F is the set of all attribute sentiment scores; S is the subset of attributes that does not contain A i , f(S) is the predicted value of the model on the attribute subset S, and f(S∪{A i}) - f(S) is the marginal contribution of attribute A i to the subset S; is the weight, which is used to ensure that the contributions of all product attribute combinations are fairly distributed.

[0125] To estimate the Shapley value in the neural network model, the SHAP method infers the weights between the input layer and the output layer by combining the impacts of the features of the neural network widgets on the entire neural network. In each trained neural network, the SHAP values of the product attributes mentioned in each comment can be obtained through calculation, as shown in Table 2:

[0126] Table 2

[0127]

[0128]

[0129] Then, in the k-th neural network model, the product attribute importance Imp ik can be obtained, and the calculation formula is as follows:

[0130]

[0131] where, n k is the number of online reviews used to train the k-th neural network, and j is the index.

[0132] Finally, taking the F1 value of each neural network as the weight, the product attribute importance measured from k neural networks is weighted and summed, and the calculation formula is as follows:

[0133]

[0134] where, w k is the F1 value of the k-th neural network.

[0135] In step S5, based on the performance of the product attributes and the importance of the product attributes, the product attributes are mapped to a two-dimensional plane within the IPA model framework to obtain the product attribute improvement strategy for the target product.

[0136] Based on the measured and Imp m , as Figure 3 shown, in this step, the product attributes of the target product are mapped to a two-dimensional plane. According to the IPA model framework, the improvement strategies of product attributes are usually divided into four categories, and the division thresholds of importance and performance need to be set. For the division threshold, the common average value is used as the threshold. The four specific product attribute improvement strategies are as follows:

[0137] (1) Continue to maintain: The importance of the product attribute is higher than the set threshold, indicating that the product attribute is very important in the consumer purchase decision; the performance of the product attribute is also higher than the set threshold, indicating that the performance of the product attribute is in a comparative advantage position in the competitive market. Such product attributes belong to the core product competitiveness of the enterprise, and the competitive advantage of the product attribute should be maintained.

[0138] (2) Focus on: The importance of the product attribute is higher than the set threshold, indicating that the product attribute is very important in the consumer purchase decision; the performance of the product attribute is lower than the set threshold, indicating that the performance of the product attribute is in a comparative disadvantage position in the competitive market. This type of product attribute is an important product defect of the enterprise, and the performance of this type of product attribute should be preferentially improved.

[0139] (3) Over-effort: The importance of the product attribute is lower than the set threshold, indicating that consumers do not attach importance to this type of product attribute in the purchase decision-making process; the performance of the product attribute is higher than the set threshold, indicating that the performance of the product attribute is higher than the average level of the competitive market. The excellent performance of this type of product attribute has little impact on the improvement of product competitiveness, and the enterprise should avoid excessive resource investment in this type of product attribute.

[0140] (4) Low priority: The importance of the product attribute is lower than the set threshold, indicating that consumers do not attach enough importance to this type of product attribute in the purchase decision-making process; the performance of the product attribute is also lower than the set threshold, indicating that the performance of the product attribute is worse than the average level of the competitive market. Since improving the performance of this type of product attribute may generate very little market economic benefits, the enterprise should place this type of product attribute in a low-priority improvement position.

[0141] So far, the embodiment of the present invention has completed all the processes of the product strategy improvement method based on LSGDM-SHAP.

[0142] Embodiment 2:

[0143] An embodiment of the present invention provides a product strategy improvement system based on LSGDM-SHAP, including:

[0144] A comment collection module, configured to collect a number of online comparison comments published by consumers after purchasing the target product or its single competing product; each of the online comparison comments includes attribute comparison information between the target product and the single competing product;

[0145] An emotion analysis module, configured to extract the attribute information in the online comparison comments and perform consumer emotion analysis on the attribute information to obtain the emotion scores of different product attributes under each purchase result;

[0146] A performance display acquisition module, configured to seek consensus among different consumer views using the LSGDM method based on the emotion scores to obtain the product attribute performance that best represents the overall view;

[0147] An attribute importance acquisition module, configured to pre-train a neural network model corresponding to each product attribute based on the emotion scores and purchase results, calculate the Shapley value of each product attribute using the SHAP interpretability method, and multiply it by the emotion score to obtain the contribution value of different product attributes to the predicted purchase result of the model, and use it as the product attribute importance;

[0148] An improvement strategy acquisition module, configured to map product attributes to a two-dimensional plane under the IPA model framework based on the product attribute performance and the product attribute importance to obtain the product attribute improvement strategy of the target product.

[0149] Embodiment 3:

[0150] An embodiment of the present invention provides a storage medium that stores a computer program for improving product strategy based on LSGDM-SHAP, wherein the computer program causes a computer to execute the product strategy improvement method as described in Embodiment 1.

[0151] Embodiment 4:

[0152] An embodiment of the present invention provides an electronic device, including:

[0153] One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include those for executing the product strategy improvement method as described in Embodiment 1.

[0154] It is understandable that the product strategy improvement system, storage medium, and electronic device provided in the embodiments of the present invention corresponding to the product strategy improvement method based on LSGDM-SHAP. For the explanations, examples, beneficial effects, and other parts of the relevant content, reference can be made to the corresponding parts in the product strategy improvement method, which will not be elaborated here.

[0155] In summary, compared with the prior art, the following beneficial effects are achieved:

[0156] In the embodiments of the present invention, first, a number of online comparison reviews published after consumers purchase the target product or its competing products are collected; second, the attribute information in the online comparison reviews is extracted and consumer sentiment analysis is performed; third, the LSGDM method is used to seek consensus among different consumer views to obtain the product attribute performance that best represents the overall view; then, the SHAP interpretability method is used to obtain the importance of product attributes; finally, the product attributes are mapped to a two-dimensional plane within the IPA model framework to obtain the product attribute improvement strategy for the target product. From the perspective of market competition, by using online comparison reviews, the consumer needs can be quickly and accurately understood, which helps enterprises grasp the key points of product improvement, provides product improvement strategies for enterprises, and enables the effective and reasonable utilization of enterprise resources.

[0157] It should be noted that in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise", or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article, or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article, or device including the said element.

[0158] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A product strategy improvement method based on LSGDM-SHAP, characterized in that: include: Collect online comparative reviews posted by several consumers after purchasing the target product or its competitors; Each of the online comparative reviews includes comparative information on attributes of the target product and competing products; Extracting attribute information from the online comparative reviews, and performing consumer sentiment analysis on the attribute information to obtain sentiment scores for different product attributes in each of the online comparative reviews; Based on the sentiment scores, the LSGDM method is used to seek consensus among different consumer opinions to obtain the product attribute performance that best represents the overall opinion; Based on the sentiment score and the purchase result, a neural network model corresponding to each product attribute is pre-trained, the Shapley value of each product attribute is calculated using the SHAP interpretability method, and the Shapley value is multiplied by the sentiment score to obtain the contribution value of different product attributes to the purchase result predicted by the model, and used as the importance of the product attribute; Based on the performance of the product attributes and the importance of the product attributes, the product attributes are mapped to a two-dimensional plane under the IPA model framework to obtain a product attribute improvement strategy for the target product.

2. The product strategy improvement method according to claim 1, characterized in that: LDA is used to extract attribute information from the online comparative reviews.

3. The product strategy improvement method according to claim 1, characterized in that: The performing consumer sentiment analysis on the attribute information includes: The conditional random field method is used to identify product names; The Snownlp tool was used for sentiment analysis, and the sentiment was expressed as a number between 0 and 1. If the score was greater than 0.5, it means that the sentiment was positive; conversely, if the score was less than 0.5, it means that the sentiment was negative.

4. The product strategy improvement method according to claim 1, characterized in that: Based on the sentiment scores, the LSGDM method is used to seek consensus among different consumer opinions to obtain the product attribute performance that best represents the overall opinion, including: Using Kmeans algorithm to cluster all online comparative reviews into multiple subgroup opinions, and calculating the comprehensive expression of each subgroup opinion; By constructing a variance minimization objective function, the opinion weight of each subgroup's opinion is continuously optimized; Based on the best opinion weights of each subgroup's views and the adjusted comprehensive expression, the weighted evaluation reaches an overall decision consensus and serves as the product attribute performance measurement result that best represents the overall viewpoint.

5. The product strategy improvement method according to claim 1, characterized in that: The SHAP interpretability method is used to calculate the Shapley value of each product attribute. The calculation formula is: in, It is attribute A i Shapley value, F is the set of all attribute sentiment scores; S is the set that does not include A i The attribute subset of A, f(S) is the predicted value of the model on the attribute subset S, f(S∪{A i })-f(S) is attribute A i The marginal contribution to the subset S; are weights used to ensure that the contributions of all product attribute combinations are distributed fairly.

6. The product strategy improvement method according to claim 1, characterized in that: The product attribute improvement strategy includes: Set the thresholds for product attribute importance and product attribute performance respectively; If both the importance of product attributes and the performance of product attributes are higher than the corresponding thresholds, the product attribute improvement strategy is to continue to maintain; If the importance of a product attribute is higher than the corresponding threshold, and the performance of a product attribute is lower than the corresponding threshold, then the product attribute improvement strategy is to focus on it; If the importance of the product attribute is lower than the corresponding threshold, and the performance of the product attribute is higher than the corresponding threshold, then the product attribute improvement strategy is over-effort; If both the product attribute importance and the product attribute performance are lower than the corresponding thresholds, the product attribute enhancement strategy is of low priority.

7. A product strategy improvement system based on LSGDM-SHAP, characterized in that: include: A review collection module is used to collect online comparative reviews posted by several consumers after they purchase the target product or a single competitive product thereof; Each of the online comparative reviews includes comparative information on attributes of the target product and a single competing product; A sentiment analysis module, used to extract attribute information from the online comparative reviews, and perform consumer sentiment analysis on the attribute information to obtain sentiment scores of different product attributes under each purchase result; A performance acquisition module, for seeking consensus among different consumer opinions by using the LSGDM method based on the sentiment score, so as to obtain the product attribute performance that best represents the overall opinion; An attribute importance acquisition module, used to pre-train a neural network model corresponding to each product attribute based on the sentiment score and purchase results, calculate the Shapley value of each product attribute using the SHAP interpretability method, and multiply it by the sentiment score to obtain the contribution value of different product attributes to the purchase results predicted by the model, and use it as the product attribute importance; The improvement strategy acquisition module is used to map the product attributes to a two-dimensional plane under the IPA model framework based on the product attribute performance and the importance of the product attributes, so as to obtain the product attribute improvement strategy of the target product.

8. A storage medium, characterized in that: The computer program for improving product strategy based on LSGDM-SHAP is stored therein, wherein the computer program enables a computer to execute the product strategy improvement method according to any one of claims 1 to 6.

9. An electronic device, characterized in that: include: one or more processors; Memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, the programs including programs for executing the product strategy improvement method according to any one of claims 1 to 6.

Images