Commodity improvement suggestion generation method and device and comment analysis system

Abstract

The invention provides a commodity improvement suggestion generation method and device and a comment analysis system, and the method comprises the steps: obtaining the feature information of a user of a target commodity, obtaining first feature information, and enabling the feature information to comprise one or more of historical behaviors, historical comments and social influence of the user on a consumption platform; forming a first feature vector of each user by the first feature information of each user; inputting each first feature vector into a credibility model to obtain the credibility of each user; filtering the comments of the users whose credibility is lower than a predetermined credibility to obtain referable comments of the target commodity; the reference comments of the target commodity are aggregated to generate the comment abstract of the target commodity, and the improvement suggestion of the target commodity is generated according to the problem feedback content of the comment abstract, so that the problem that the improvement suggestion obtained by comment analysis cannot improve the user satisfaction degree due to distortion of online comments caused by subjective factors of a user in the prior art is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of comment analysis, in particular to a generation method and device of commodity improvement suggestions, a computer readable storage medium and a comment analysis system. BACKGROUND

[0002] The existing user portrait technology relies on static features (such as historical behavior), and it is difficult to dynamically capture the real-time intention and emotional tendency of the user when publishing comments, resulting in a disconnection between the portrait and the authenticity of the comments. The reason is that the traditional portrait construction focuses on long-term behavior statistics (such as consumption records), ignores short-term emotional fluctuations and motivation changes in the comment scenario, and cannot identify temporary bias or malicious distortion.

[0003] Moreover, the comment extraction and summary method is usually based on text similarity aggregation, and does not fuse the user portrait credibility weight, which can easily amplify the spread of low-quality comments. The reason is that the existing summary algorithm is content-centered, and does not establish a quantitative correlation chain of "user credibility-comment value", which cannot filter distorted content caused by personal emotions and interest-driven. SUMMARY

[0004] The main purpose of the present application is to provide a generation method, device, computer readable storage medium and comment analysis system of commodity improvement suggestions, so as to at least solve the problem that the online comments are distorted due to user subjective factors in the prior art, and the improvement suggestions obtained by comment analysis cannot improve user satisfaction.

[0005] In order to achieve the above purpose, according to one aspect of the present application, a generation method of commodity improvement suggestions is provided, comprising: obtaining the feature information of the user of the target commodity to obtain the first feature information, the feature information comprising one or more of the historical behavior, the historical comment and the social influence of the user on the consumption platform, the historical behavior comprising the consumption frequency, and the social influence comprising the number of fans and the interaction rate; the first feature information of each user is combined to form the first feature vector of each user; each first feature vector is input into a credibility model to obtain the credibility of each user, the credibility model being a model for analyzing credibility trained by reinforcement learning or genetic algorithm; filtering out the comments of the user whose credibility is lower than the predetermined credibility to obtain the referenceable comments of the target commodity; aggregating the referenceable comments of the target commodity to generate a comment summary of the target commodity, and generating an improvement suggestion of the target commodity according to the problem feedback content of the comment summary.

[0006] Optionally, before inputting the feature vectors of each user into the credibility model to obtain the credibility of each user, the method further includes: a first acquisition step, acquiring feature information of other users to obtain second feature information, wherein the other users are users of other products; a combination step, combining the second feature information of each of the other users to form a second feature vector of each of the other users; an analysis step, inputting each of the second feature vectors into an initial credibility model to obtain the credibility of each of the other users; a filtering step, filtering out comments of the other users whose credibility is lower than the predetermined credibility to obtain referenceable comments of the other products; a first generation step, aggregating and generating a comment summary of the other products based on the referenceable comments of the other products; and a second acquisition step, acquiring the click-through rate and ranking of the comment summary of the other products. The report rate is defined as follows: the click-through rate is the percentage of users who clicked to view the comment summary among those browsing other products; the report rate is the percentage of users who reported the comment summary as inaccurate among those who clicked to view the comment summary. The calculation steps involve multiplying the click-through rate by its corresponding weight to obtain a positive reward, multiplying the report rate by its corresponding weight to obtain a negative reward, and calculating the difference between the positive and negative rewards to obtain the reward value of the comment summary. Each time the parameters of the initial credibility model are adjusted, the first acquisition step, the combination step, the analysis step, the filtering step, the first generation step, the second acquisition step, and the calculation step are repeated sequentially until the maximum reward value of the comment summary is reached. The initial credibility model corresponding to the maximum reward value is then determined as the credibility model.

[0007] Optionally, before inputting the feature vectors of each user into the credibility model to obtain the credibility of each user, the method further includes: a second generation step, randomly generating multiple model parameter sets to obtain an initial population, wherein the model parameter sets include all the parameters to be optimized in the initial credibility model; a determination step, determining the reward value of the comment summary corresponding to the model parameter set as the fitness of the individual in the corresponding initial population; an adjustment step, determining the individuals with fitness greater than a predetermined fitness as high-quality individuals, performing cross-operation on multiple high-quality individuals to obtain multiple offspring individuals, and randomly adjusting the parameters of at least one offspring individual to obtain at least one mutated individual; a replacement step, replacing the individuals with fitness lower than the predetermined fitness with the mutated individual and the unmutated offspring individuals to obtain an updated population; repeating the second generation step, the determination step, the adjustment step, and the replacement step a predetermined number of times in sequence to obtain a latest population; determining the model parameter set corresponding to the individual with the highest fitness in the latest population as the optimal model parameter set; and determining the initial credibility model corresponding to the optimal model parameter set as the credibility model.

[0008] Optionally, inputting each of the first feature vectors into the credibility model to obtain the credibility of each user includes: using a sparse autoencoder in the credibility model to reduce the dimensionality of each of the first feature vectors to obtain multiple first compressed feature vectors; and inputting each of the first compressed feature vectors into the GBDT-Transformer fusion model in the credibility model to obtain the credibility of each user.

[0009] Optionally, the referenced reviews of the target product are aggregated to generate a review summary of the target product, including: calculating the weight of each referenced review based on the credibility of all the referenced reviews and the corresponding users; retaining the referenced review with the highest weight when multiple referenced reviews have the same semantic substance; and arranging the remaining referenced reviews in descending order of weight to obtain the review summary.

[0010] Optionally, the weight of each referenced comment is calculated based on the credibility of all the referenced comments and the corresponding users, including: calculating the sum of the credibility of the users corresponding to all the referenced comments to obtain the cumulative credibility; calculating the ratio of the credibility of the users corresponding to each referenced comment to the cumulative credibility to obtain the credibility weight of each referenced comment; converting each referenced comment into a feature vector to obtain multiple comment feature vectors; calculating the average of the multiple comment feature vectors to obtain an average feature vector; calculating the cosine similarity between each comment feature vector and the average feature vector to obtain the similarity weight of each referenced comment; and calculating the product of the credibility weight and the similarity weight of each referenced comment to obtain the weight of each referenced comment.

[0011] Optionally, before filtering out comments from users whose credibility is lower than a predetermined credibility level to obtain referable comments for the target product, the method further includes: updating the credibility of the users for the target product at predetermined intervals.

[0012] According to another aspect of this application, an apparatus for generating product improvement suggestions is provided, comprising: a first acquisition unit, configured to acquire user feature information of a target product to obtain first feature information, the feature information including one or more of the user's historical behavior, historical comments, and social influence on a consumption platform, the historical behavior including consumption frequency, and the social influence including the number of followers and interaction rate; a first combination unit, configured to combine the first feature information of each user to form a first feature vector of each user; a first analysis unit, configured to input each first feature vector into a credibility model to obtain the credibility of each user, the credibility model being a credibility analysis model trained by reinforcement learning or a genetic algorithm; a first filtering unit, configured to filter out comments of the users whose credibility is lower than a predetermined credibility to obtain referable comments of the target product; and a first generation unit, configured to aggregate the referable comments of the target product to generate a comment summary of the target product, and generate improvement suggestions for the target product based on the problem feedback content of the comment summary.

[0013] According to another aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform any of the aforementioned methods for generating product improvement suggestions.

[0014] According to another aspect of this application, a review analysis system is provided, comprising: one or more processors, a memory, and one or more programs, wherein the review analysis system is applied to a consumer platform, the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include a method for generating any of the product improvement suggestions described above.

[0015] By applying the technical solution of this application, the above-mentioned method for generating product improvement suggestions analyzes the credibility of users by using the user characteristic information of the target product to filter out comments from users whose credibility is lower than a predetermined credibility level. Based on the remaining referable comments, improvement suggestions for the target product can be generated, ensuring that the referable comments can truly reflect the quality of the target product. This generates improvement suggestions that can improve user satisfaction, allowing manufacturers to improve the target product based on the improvement suggestions to increase user satisfaction. This solves the problem in the prior art where online comments are distorted due to user subjective factors, making the improvement suggestions obtained from comment analysis unable to improve user satisfaction. Attached Figure Description

[0016] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0017] Figure 1 A hardware structure block diagram of a mobile terminal for generating product improvement suggestions according to an embodiment of this application is shown.

[0018] Figure 2 A flowchart illustrating a method for generating product improvement suggestions according to an embodiment of this application is shown.

[0019] Figure 3 A structural block diagram of an apparatus for generating product improvement suggestions according to an embodiment of this application is shown.

[0020] The above figures include the following reference numerals:

[0021] 102. Processor; 104. Memory; 106. Transmission device; 108. Input / output device. Detailed Implementation

[0022] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0023] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0024] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] As described in the background section, existing online reviews suffer from distortion due to user subjectivity, making it impossible for improvement suggestions obtained from review analysis to enhance user satisfaction. To address this technical problem, embodiments of this application provide a method, apparatus, computer-readable storage medium, and review analysis system for generating product improvement suggestions.

[0026] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0027] The methods and embodiments provided in this application can be executed on a mobile terminal, computer terminal, or similar computing device. Taking running on a mobile terminal as an example, Figure 1 This is a hardware structure block diagram of a mobile terminal for a product improvement suggestion generation method according to an embodiment of the present invention. Figure 1 As shown, a mobile terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 104 for storing data are also shown. The mobile terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the mobile terminal described above. For example, the mobile terminal may also include components that are more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0028] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the product improvement suggestion generation method in this embodiment of the invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, thereby implementing the above-described method. The memory 104 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. The transmission device 106 is used to receive or send data via a network. Specific examples of the aforementioned networks may include wireless networks provided by the mobile terminal's communication provider. In one example, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0029] This embodiment provides a method for generating product improvement suggestions that run on a mobile terminal, computer terminal, or similar computing device. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Also, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0030] Figure 2 This is a flowchart of a method for generating product improvement suggestions according to an embodiment of this application. For example... Figure 2 As shown, the method includes the following steps:

[0031] Step S201: Obtain the characteristic information of the user of the target product to obtain the first characteristic information. The characteristic information includes one or more of the user's historical behavior, historical comments and social influence on the consumption platform. The historical behavior includes the frequency of consumption and the social influence includes the number of fans and the interaction rate.

[0032] Specifically, user characteristics include three categories: historical behavior (purchase frequency, device type), historical comments (text / rating), and social influence (number of followers, interaction rate), in order to generate user profiles.

[0033] Step S202: The first feature information of each user is used to form the first feature vector of each user.

[0034] Specifically, the three types of features mentioned above are concatenated to form the first feature vector of each user, breaking through the limitations of static profiles and introducing cross-modal correlation analysis of "emotion-behavior-influence". Among them, historical comments can be used to analyze user emotions.

[0035] Step S203: Input each of the above first feature vectors into the credibility model to obtain the credibility of each of the above users. The credibility model is a credibility analysis model obtained by training through reinforcement learning or genetic algorithm.

[0036] Specifically, the credibility analysis model, trained through reinforcement learning or genetic algorithms, can update the user credibility in real time within the range of 0-1 by updating the first feature vector.

[0037] Step S204: Filter out the comments from the users whose credibility is lower than the predetermined credibility, and obtain the reference comments for the target product.

[0038] Specifically, low-scoring user reviews are automatically filtered out based on a profile credibility threshold, such as those with a credibility score of less than 0.3.

[0039] Step S205: Aggregate the referenced reviews of the target product to generate a review summary of the target product, and generate improvement suggestions for the target product based on the feedback content of the review summary.

[0040] Specifically, the retained, referable reviews are aggregated by weight, for example, using TF-IDF and BERT semantic clustering, to generate review summaries and store them in Elasticsearch. Based on the feedback content of the review summaries, improvement suggestions for the target product can be generated. For example, based on feedback about short battery life, improvement suggestions for increasing battery capacity can be generated.

[0041] In this embodiment, the method for generating product improvement suggestions analyzes user credibility by using the user's characteristic information about the target product to filter out comments from users whose credibility is lower than a predetermined level. Based on the remaining referable comments, improvement suggestions for the target product are generated, ensuring that the referable comments accurately reflect the quality of the target product. This generates improvement suggestions that can improve user satisfaction, allowing manufacturers to improve the target product and increase user satisfaction. This solves the problem in the prior art where online comments are distorted due to user subjectivity, preventing improvement suggestions obtained from comment analysis from improving user satisfaction.

[0042] To ensure the accuracy of the credibility, in an optional implementation, before inputting the feature vectors of each user into the credibility model to obtain the credibility of each user, the method further includes:

[0043] Step S301, the first acquisition step, acquire the feature information of other users to obtain the second feature information, wherein the aforementioned other users are users of other products;

[0044] Step S302, combining steps, to form the second feature information of each of the other users into the second feature vector of each of the other users;

[0045] Step S303, analysis step: input each of the above-mentioned second feature vectors into the initial credibility model to obtain the credibility of each of the above-mentioned other users;

[0046] Step S304, filtering step, filtering out the comments of the other users whose credibility is lower than the predetermined credibility, and obtaining the reference comments of the other products;

[0047] Step S305, the first generation step, generates a summary of reviews for the other products based on the above-mentioned aggregated reviews of the other products;

[0048] Step S306, the second acquisition step, acquire the click-through rate and report rate of the review summaries of the other products mentioned above. The click-through rate is the percentage of users who clicked to view the review summaries among the users who browsed the other products mentioned above. The report rate is the percentage of users who reported the review summaries as untrue among the users who clicked to view the review summaries.

[0049] Step S307, Calculation step: Calculate the product of the above click rate and the corresponding weight to obtain the positive reward; calculate the product of the above report rate and the corresponding weight to obtain the negative reward; calculate the difference between the above positive reward and the above negative reward to obtain the reward value of the above comment summary.

[0050] Step S308: Each time the parameters of the initial credibility model are adjusted, repeat the first acquisition step, the combination step, the analysis step, the filtering step, the first generation step, the second acquisition step, and the calculation step once in sequence until the maximum reward value of the comment summary is reached. Then, determine the initial credibility model corresponding to the maximum reward value as the credibility model.

[0051] In the above implementation, the initial credibility model is trained using the feature information of other users. The reward function is R = 0.6 × click rate - 0.4 × report rate. The state space is {report rate, summary click rate}. The action space is {threshold increment / decrement step size 0.05, weight coefficient ±0.1}. These are the optimal parameters of the credibility model. After training, the credibility model is obtained, and the accurate credibility of users can be analyzed.

[0052] To ensure the accuracy of the credibility, in an optional implementation, before inputting the feature vectors of each user into the credibility model to obtain the credibility of each user, the method further includes:

[0053] Step S401, the second generation step, randomly generate multiple model parameter sets to obtain an initial population. The aforementioned model parameter sets include all the parameters to be optimized in the initial confidence model.

[0054] Step S402, Determine the step by determining the reward value of the above comment summary corresponding to the above model parameter group as the fitness of the individual in the above initial population.

[0055] Step S403, adjustment step: identify the individuals with fitness greater than the predetermined fitness as high-quality individuals, perform crossover operation on multiple high-quality individuals to obtain multiple offspring individuals, and randomly adjust the parameters of at least one offspring individual to obtain at least one mutant individual.

[0056] Step S404, replacement step, replacing the individuals with fitness lower than the predetermined fitness with the above-mentioned mutated individuals and the above-mentioned unmutated offspring individuals to obtain an updated population;

[0057] Step S405: Repeat the above second generation step, the above determination step, the above adjustment step, and the above replacement step a predetermined number of times to obtain the latest population;

[0058] Step S406: Determine the model parameter set corresponding to the individual with the highest fitness in the latest population as the optimal model parameter set;

[0059] Step S407: Determine the initial credibility model corresponding to the above-mentioned optimal model parameter set as the above-mentioned credibility model.

[0060] In the above implementation, the reinforcement learning module can be replaced by a genetic algorithm with a population size of 50, a crossover probability of 0.8, and a mutation probability of 0.05. This is suitable for vertical domains with stable rules, and can train an accurate credibility model to analyze and obtain accurate user credibility.

[0061] To achieve cross-modal correlation analysis of "emotion-behavior-influence", in one optional implementation, step S203 above includes:

[0062] Step S2031: Use the sparse autoencoder in the above credibility model to reduce the dimensionality of each of the first feature vectors to obtain multiple first compressed feature vectors.

[0063] Step S2032: Input each of the above-mentioned first compressed feature vectors into the GBDT-Transformer fusion model in the above-mentioned credibility model to obtain the credibility of each of the above-mentioned users.

[0064] In the above implementation, the GBDT-Transformer fusion model includes a GBDT part and a Transformer part. The GBDT part has a tree depth of 8 and a feature sampling rate of 0.7. The Transformer part has 4 heads, a hidden layer dimension of 128, and Dropout of 0.2. The real-time profile update process dynamically consumes incremental data from Kafka (including user behavior logs, real-time comment sentiment analysis, and diverse temporal features such as social interactions), driving feature engineering to construct a 256-dimensional feature vector every 5 minutes. A sparse autoencoder performs dimensionality reduction and compression on the features (256 → 32 dimensions), retaining the core patterns while removing noise. The GBDT-Transformer fusion model calculates credibility through a dual-path approach: the GBDT branch parses feature cross-relationships (e.g., "high consumption frequency + low sentiment fluctuation"), and the Transformer branch captures temporal patterns of comment sentiment (e.g., sudden clusters of negative reviews). The model ultimately outputs a dynamic credibility score in the 0-1 range, updating the Redis profile cache.

[0065] To achieve semantic clustering, in one optional implementation, step S205 above includes:

[0066] Step S2051: Calculate the weight of each of the aforementioned reference comments based on all the aforementioned reference comments and the credibility of the corresponding users;

[0067] Step S2052: If multiple referenced comments have the same semantic substance, retain the referenced comment with the highest weight.

[0068] Step S2053: Arrange the remaining reference comments in descending order of their weights to obtain the comment summary.

[0069] In the above implementation, comments from users with medium to high credibility are assigned weights according to their scores, semantic analysis is performed on the reference comments to remove duplicates, and the remaining reference comments are arranged in descending order of their weights to obtain the above comment summary.

[0070] In an optional implementation, to calculate the weight of the referable comments, step S2051 above includes:

[0071] Step S20511: Calculate the sum of the credibility of all the users corresponding to the above-mentioned referable comments to obtain the cumulative credibility.

[0072] Step S20512: Calculate the ratio of the credibility of the user corresponding to each of the above-mentioned reference comments to the cumulative credibility of the above-mentioned reference comments, and obtain the credibility weight of each of the above-mentioned reference comments;

[0073] Step S20513: Convert each of the above-mentioned reference comments into feature vectors to obtain multiple comment feature vectors;

[0074] Step S20514: Calculate the average value of multiple comment feature vectors to obtain the average feature vector;

[0075] Step S20515: Calculate the cosine similarity between each of the above-mentioned comment feature vectors and the above-mentioned average feature vector to obtain the similarity weight of each of the above-mentioned reference comments;

[0076] Step S20516: Calculate the product of the credibility weight and the similarity weight of each of the above-mentioned reference comments to obtain the weight of each of the above-mentioned reference comments.

[0077] In the above embodiments, the weighted formula of the aforementioned reference comments is as follows: Where credibility i is the credibility of the i-th user, C i C represents the feature vector of a comment that can be referenced. avg Let W be the average feature vector of the referenced comments, n be the number of referenced comments, and W be the number of referenced comments. i The weighting of the above-mentioned comments is used for reference.

[0078] To generate dynamic user profiles, in one optional implementation, before filtering out comments from users whose credibility is lower than a predetermined credibility level to obtain referable reviews of the target product, the method further includes:

[0079] Step S501: Update the credibility of the aforementioned user for the aforementioned target product at predetermined intervals.

[0080] In the above implementation, by updating the user's characteristic information of the target product once every predetermined time interval (5 minutes), the credibility of the user of the target product can be updated once through credibility analysis, thereby generating a dynamic user profile.

[0081] To enable those skilled in the art to better understand the technical solution of this application, the implementation process of the product improvement suggestion generation method of this application will be described in detail below with reference to specific embodiments.

[0082] This application also provides an apparatus for generating product improvement suggestions. It should be noted that this apparatus can be used to execute the product improvement suggestion generation method provided in this application. This apparatus is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0083] The following describes the apparatus for generating product improvement suggestions provided in the embodiments of this application.

[0084] Figure 3 This is a schematic diagram of an apparatus for generating product improvement suggestions according to an embodiment of this application. Figure 3 As shown, the device includes:

[0085] The first acquisition unit 10 is used to acquire the user's feature information of the target product and obtain the first feature information. The feature information includes one or more of the user's historical behavior, historical comments and social influence on the consumption platform. The historical behavior includes consumption frequency and the social influence includes the number of fans and the interaction rate.

[0086] Specifically, user characteristics include three categories: historical behavior (purchase frequency, device type), historical comments (text / rating), and social influence (number of followers, interaction rate), in order to generate user profiles.

[0087] The first combination unit 20 is used to combine the first feature information of each of the above-mentioned users into a first feature vector of each of the above-mentioned users.

[0088] Specifically, the three types of features mentioned above are concatenated to form the first feature vector of each user, breaking through the limitations of static profiles and introducing cross-modal correlation analysis of "emotion-behavior-influence". Among them, historical comments can be used to analyze user emotions.

[0089] The first analysis unit 30 is used to input each of the aforementioned first feature vectors into the credibility model to obtain the credibility of each of the aforementioned users. The credibility model is an analysis credibility model trained by reinforcement learning or genetic algorithm.

[0090] Specifically, the credibility analysis model, trained through reinforcement learning or genetic algorithms, can update the user credibility in real time within the range of 0-1 by updating the first feature vector.

[0091] The first filtering unit 40 is used to filter out the comments of the users whose credibility is lower than the predetermined credibility, and obtain the reference comments of the target product.

[0092] Specifically, low-scoring user reviews are automatically filtered out based on a profile credibility threshold, such as those with a credibility score of less than 0.3.

[0093] The first generation unit 50 is used to aggregate the aforementioned referable reviews of the target product to generate a review summary of the target product, and to generate improvement suggestions for the target product based on the problem feedback content of the review summary.

[0094] Specifically, the retained, referable reviews are aggregated by weight, for example, using TF-IDF and BERT semantic clustering, to generate review summaries and store them in Elasticsearch. Based on the feedback content of the review summaries, improvement suggestions for the target product can be generated. For example, based on feedback about short battery life, improvement suggestions for increasing battery capacity can be generated.

[0095] In this embodiment, the above-mentioned product improvement suggestion generation device analyzes the credibility of users by using the user's characteristic information of the target product to filter out comments from users whose credibility is lower than a predetermined credibility level. Based on the remaining referable comments, improvement suggestions for the target product can be generated, ensuring that the referable comments can truly reflect the quality of the target product. This generates improvement suggestions that can improve user satisfaction, allowing manufacturers to improve the target product based on the improvement suggestions to increase user satisfaction. This solves the problem in the prior art where online comments are distorted due to user subjective factors, making the improvement suggestions obtained from comment analysis unable to improve user satisfaction.

[0096] To ensure the accuracy of the reliability, in an optional embodiment, the above-mentioned device further includes:

[0097] The second acquisition unit is used to perform the first acquisition step. Before inputting the feature vectors of each of the above-mentioned users into the credibility model to obtain the credibility of each of the above-mentioned users, the unit acquires the feature information of other users to obtain the second feature information. The above-mentioned other users are users of other products.

[0098] The second combination unit is used to perform the combination step, which combines the second feature information of each of the other users into the second feature vector of each of the other users.

[0099] The second analysis unit is used to perform the analysis steps, inputting each of the above-mentioned second feature vectors into the initial credibility model to obtain the credibility of each of the above-mentioned other users;

[0100] The second filtering unit is used to perform the filtering step, filtering out the comments of the other users whose credibility is lower than the predetermined credibility, and obtaining the reference comments of the other products.

[0101] The second generation unit is used to perform the first generation step, generating a summary of reviews for the other products based on the aggregated reviews of the other products mentioned above.

[0102] The third acquisition unit is used to perform the second acquisition step to acquire the click-through rate and report rate of the review summaries of the other products mentioned above. The click-through rate is the percentage of users who clicked to view the review summaries among the users who browsed the other products mentioned above. The report rate is the percentage of users who reported the review summaries as untrue among the users who clicked to view the review summaries.

[0103] The calculation unit is used to calculate the product of the above click rate and the corresponding weight to obtain a positive reward, calculate the product of the above report rate and the corresponding weight to obtain a negative reward, and calculate the difference between the above positive reward and the above negative reward to obtain the reward value of the above comment summary.

[0104] The first adjustment unit is used to repeat the first acquisition step, the combination step, the analysis step, the filtering step, the first generation step, the second acquisition step, and the calculation step once in sequence for each adjustment of the parameters of the initial credibility model, until the maximum reward value of the comment summary is reached, and the initial credibility model corresponding to the maximum reward value is determined as the credibility model.

[0105] In the above implementation, the initial credibility model is trained using the feature information of other users. The reward function is R = 0.6 × click rate - 0.4 × report rate. The state space is {report rate, summary click rate}. The action space is {threshold increment / decrement step size 0.05, weight coefficient ±0.1}. These are the optimal parameters of the credibility model. After training, the credibility model is obtained, and the accurate credibility of users can be analyzed.

[0106] To ensure the accuracy of the reliability, in an optional embodiment, the above-mentioned device further includes:

[0107] The third generation unit is used to perform the second generation step. Before inputting the feature vectors of each user into the credibility model to obtain the credibility of each user, it randomly generates multiple model parameter sets to obtain an initial population. The model parameter sets include all the parameters to be optimized in the initial credibility model.

[0108] The first determining unit is used to perform the determining step, which determines the reward value of the above comment summary corresponding to the above model parameter set as the fitness of the individual in the above initial population.

[0109] The second adjustment unit is used to perform the adjustment steps, identify the individuals with fitness greater than a predetermined fitness as superior individuals, perform crossover operations on multiple superior individuals to obtain multiple offspring individuals, and randomly adjust the parameters of at least one offspring individual to obtain at least one mutated individual.

[0110] A replacement unit is used to perform a replacement step, replacing the individuals with fitness lower than the predetermined fitness with the aforementioned mutated individuals and the aforementioned unmutated offspring individuals to obtain an updated population;

[0111] The repeating unit is used to sequentially repeat the second generation step, the determination step, the adjustment step, and the replacement step a predetermined number of times to obtain the latest population.

[0112] The second determining unit is used to determine the model parameter set corresponding to the individual with the highest fitness in the latest population as the optimal model parameter set.

[0113] The third determining unit is used to determine the initial credibility model corresponding to the above-mentioned optimal model parameter set as the above-mentioned credibility model.

[0114] In the above implementation, the reinforcement learning module can be replaced by a genetic algorithm with a population size of 50, a crossover probability of 0.8, and a mutation probability of 0.05. This is suitable for vertical domains with stable rules, and can train an accurate credibility model to analyze and obtain accurate user credibility.

[0115] To achieve cross-modal correlation analysis of "emotion-behavior-influence", in one optional implementation, the first analysis unit mentioned above includes:

[0116] The processing module is used to reduce the dimensionality of each of the first feature vectors by using the sparse autoencoder in the above credibility model to obtain multiple first compressed feature vectors.

[0117] The analysis module is used to input each of the above-mentioned first compressed feature vectors into the GBDT-Transformer fusion model in the above-mentioned credibility model to obtain the credibility of each of the above-mentioned users.

[0118] In the above implementation, the GBDT-Transformer fusion model includes a GBDT part and a Transformer part. The GBDT part has a tree depth of 8 and a feature sampling rate of 0.7. The Transformer part has 4 heads, a hidden layer dimension of 128, and Dropout of 0.2. The real-time profile update process dynamically consumes incremental data from Kafka (including user behavior logs, real-time comment sentiment analysis, and diverse temporal features such as social interactions), driving feature engineering to construct a 256-dimensional feature vector every 5 minutes. A sparse autoencoder performs dimensionality reduction and compression on the features (256 → 32 dimensions), retaining the core patterns while removing noise. The GBDT-Transformer fusion model calculates credibility through a dual-path approach: the GBDT branch parses feature cross-relationships (e.g., "high consumption frequency + low sentiment fluctuation"), and the Transformer branch captures temporal patterns of comment sentiment (e.g., sudden clusters of negative reviews). The model ultimately outputs a dynamic credibility score in the 0-1 range, updating the Redis profile cache.

[0119] To achieve semantic clustering, in one optional implementation, the first generation unit includes:

[0120] The calculation module is used to calculate the weight of each of the aforementioned reference comments based on all the aforementioned reference comments and the credibility of the corresponding users;

[0121] The deletion module is used to retain the referenced comment with the highest weight when multiple referenced comments have the same semantic substance.

[0122] The sorting module is used to sort the remaining referenced comments in descending order of their weights to obtain the comment summary.

[0123] In the above implementation, comments from users with medium to high credibility are assigned weights according to their scores, semantic analysis is performed on the reference comments to remove duplicates, and the remaining reference comments are arranged in descending order of their weights to obtain the above comment summary.

[0124] In one optional implementation, to calculate the weight of the referable comments, the calculation module includes:

[0125] The first calculation submodule is used to calculate the sum of the credibility of all the aforementioned users corresponding to the above-mentioned referable comments, and obtain the cumulative credibility.

[0126] The second calculation submodule is used to calculate the ratio of the credibility of the user corresponding to each of the above-mentioned reference comments to the cumulative credibility of the above-mentioned reference comments, so as to obtain the credibility weight of each of the above-mentioned reference comments.

[0127] The third calculation submodule is used to convert each of the above-mentioned reference comments into feature vectors to obtain multiple comment feature vectors;

[0128] The fourth calculation submodule is used to calculate the average value of multiple comment feature vectors mentioned above, and obtain the average feature vector;

[0129] The fifth calculation submodule is used to calculate the cosine similarity between each of the above-mentioned comment feature vectors and the above-mentioned average feature vector, and to obtain the similarity weight of each of the above-mentioned reference comments.

[0130] Step S20516: Calculate the product of the credibility weight and the similarity weight of each of the above-mentioned reference comments to obtain the weight of each of the above-mentioned reference comments.

[0131] In the above embodiments, the weighted formula of the aforementioned reference comments is as follows: Where credibility i is the credibility of the i-th user, C i C represents the feature vector of a comment that can be referenced. avg Let W be the average feature vector of the referenced comments, n be the number of referenced comments, and W be the number of referenced comments. i The weighting of the above-mentioned comments is used for reference.

[0132] To generate dynamic user profiles, in one optional implementation, the above-mentioned apparatus further includes:

[0133] The update unit is used to update the credibility of the users of the target product once at predetermined intervals before filtering out the comments of the users whose credibility is lower than a predetermined credibility and obtaining referable comments of the target product.

[0134] In the above implementation, by updating the user's characteristic information of the target product once every predetermined time interval (5 minutes), the credibility of the user of the target product can be updated once through credibility analysis, thereby generating a dynamic user profile.

[0135] The device for generating the aforementioned product improvement suggestions includes a processor and a memory. The first acquisition unit, first combination unit, first analysis unit, first filtering unit, and first generation unit are all stored as program units in the memory. The processor executes these program units stored in the memory to achieve the corresponding functions. All of the above modules are located in the same processor; alternatively, the modules may be located in different processors in any combination.

[0136] The processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and adjusting kernel parameters can address the problem in existing technologies where online reviews are distorted due to user subjectivity, preventing improvement suggestions derived from review analysis from enhancing user satisfaction.

[0137] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0138] This invention provides a computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device containing the computer-readable storage medium to perform the method for generating the product improvement proposal.

[0139] Specifically, the methods for generating product improvement suggestions include:

[0140] Step S201: Obtain the characteristic information of the user of the target product to obtain the first characteristic information. The characteristic information includes one or more of the user's historical behavior, historical comments and social influence on the consumption platform. The historical behavior includes the frequency of consumption and the social influence includes the number of fans and the interaction rate.

[0141] Specifically, user characteristics include three categories: historical behavior (purchase frequency, device type), historical comments (text / rating), and social influence (number of followers, interaction rate), in order to generate user profiles.

[0142] Step S202: The first feature information of each user is used to form the first feature vector of each user.

[0143] Specifically, the three types of features mentioned above are concatenated to form the first feature vector of each user, breaking through the limitations of static profiles and introducing cross-modal correlation analysis of "emotion-behavior-influence". Among them, historical comments can be used to analyze user emotions.

[0144] Step S203: Input each of the above first feature vectors into the credibility model to obtain the credibility of each of the above users. The credibility model is a credibility analysis model obtained by training through reinforcement learning or genetic algorithm.

[0145] Specifically, the credibility analysis model, trained through reinforcement learning or genetic algorithms, can update the user credibility in real time within the range of 0-1 by updating the first feature vector.

[0146] Step S204: Filter out the comments from the users whose credibility is lower than the predetermined credibility, and obtain the reference comments for the target product.

[0147] Specifically, low-scoring user reviews are automatically filtered out based on a profile credibility threshold, such as those with a credibility score of less than 0.3.

[0148] Step S205: Aggregate the referenced reviews of the target product to generate a review summary of the target product, and generate improvement suggestions for the target product based on the feedback content of the review summary.

[0149] Specifically, the retained, referable reviews are aggregated by weight, for example, using TF-IDF and BERT semantic clustering, to generate review summaries and store them in Elasticsearch. Based on the feedback content of the review summaries, improvement suggestions for the target product can be generated. For example, based on feedback about short battery life, improvement suggestions for increasing battery capacity can be generated.

[0150] This invention provides a processor for running a program, wherein the program executes the method for generating the product improvement suggestions.

[0151] Specifically, the methods for generating product improvement suggestions include:

[0152] Step S201: Obtain the characteristic information of the user of the target product to obtain the first characteristic information. The characteristic information includes one or more of the user's historical behavior, historical comments and social influence on the consumption platform. The historical behavior includes the frequency of consumption and the social influence includes the number of fans and the interaction rate.

[0153] Specifically, user characteristics include three categories: historical behavior (purchase frequency, device type), historical comments (text / rating), and social influence (number of followers, interaction rate), in order to generate user profiles.

[0154] Step S202: The first feature information of each user is used to form the first feature vector of each user.

[0155] Specifically, the three types of features mentioned above are concatenated to form the first feature vector of each user, breaking through the limitations of static profiles and introducing cross-modal correlation analysis of "emotion-behavior-influence". Among them, historical comments can be used to analyze user emotions.

[0156] Step S203: Input each of the above first feature vectors into the credibility model to obtain the credibility of each of the above users. The credibility model is a credibility analysis model obtained by training through reinforcement learning or genetic algorithm.

[0157] Specifically, the credibility analysis model, trained through reinforcement learning or genetic algorithms, can update the user credibility in real time within the range of 0-1 by updating the first feature vector.

[0158] Step S204: Filter out the comments from the users whose credibility is lower than the predetermined credibility, and obtain the reference comments for the target product.

[0159] Specifically, low-scoring user reviews are automatically filtered out based on a profile credibility threshold, such as those with a credibility score of less than 0.3.

[0160] Step S205: Aggregate the referenced reviews of the target product to generate a review summary of the target product, and generate improvement suggestions for the target product based on the feedback content of the review summary.

[0161] Specifically, the retained, referable reviews are aggregated by weight, for example, using TF-IDF and BERT semantic clustering, to generate review summaries and store them in Elasticsearch. Based on the feedback content of the review summaries, improvement suggestions for the target product can be generated. For example, based on feedback about short battery life, improvement suggestions for increasing battery capacity can be generated.

[0162] This invention provides a comment analysis system applied to a consumer platform. The comment analysis system includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs at least the following steps:

[0163] Step S201: Obtain the characteristic information of the user of the target product to obtain the first characteristic information. The characteristic information includes one or more of the user's historical behavior, historical comments and social influence on the consumption platform. The historical behavior includes the frequency of consumption and the social influence includes the number of fans and the interaction rate.

[0164] Specifically, user characteristics include three categories: historical behavior (purchase frequency, device type), historical comments (text / rating), and social influence (number of followers, interaction rate), in order to generate user profiles.

[0165] Step S202: The first feature information of each user is used to form the first feature vector of each user.

[0166] Specifically, the three types of features mentioned above are concatenated to form the first feature vector of each user, breaking through the limitations of static profiles and introducing cross-modal correlation analysis of "emotion-behavior-influence". Among them, historical comments can be used to analyze user emotions.

[0167] Step S203: Input each of the above first feature vectors into the credibility model to obtain the credibility of each of the above users. The credibility model is a credibility analysis model obtained by training through reinforcement learning or genetic algorithm.

[0168] Specifically, the credibility analysis model, trained through reinforcement learning or genetic algorithms, can update the user credibility in real time within the range of 0-1 by updating the first feature vector.

[0169] Step S204: Filter out the comments from the users whose credibility is lower than the predetermined credibility, and obtain the reference comments for the target product.

[0170] Specifically, low-scoring user reviews are automatically filtered out based on a profile credibility threshold, such as those with a credibility score of less than 0.3.

[0171] Step S205: Aggregate the referenced reviews of the target product to generate a review summary of the target product, and generate improvement suggestions for the target product based on the feedback content of the review summary.

[0172] Specifically, the retained, referable reviews are aggregated by weight, for example, using TF-IDF and BERT semantic clustering, to generate review summaries and store them in Elasticsearch. Based on the feedback content of the review summaries, improvement suggestions for the target product can be generated. For example, based on feedback about short battery life, improvement suggestions for increasing battery capacity can be generated.

[0173] The devices mentioned in this article can be servers, PCs, tablets, mobile phones, etc.

[0174] This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program having at least the following method steps:

[0175] Step S201: Obtain the characteristic information of the user of the target product to obtain the first characteristic information. The characteristic information includes one or more of the user's historical behavior, historical comments and social influence on the consumption platform. The historical behavior includes the frequency of consumption and the social influence includes the number of fans and the interaction rate.

[0176] Specifically, user characteristics include three categories: historical behavior (purchase frequency, device type), historical comments (text / rating), and social influence (number of followers, interaction rate), in order to generate user profiles.

[0177] Step S202: The first feature information of each user is used to form the first feature vector of each user.

[0178] Specifically, the three types of features mentioned above are concatenated to form the first feature vector of each user, breaking through the limitations of static profiles and introducing cross-modal correlation analysis of "emotion-behavior-influence". Among them, historical comments can be used to analyze user emotions.

[0179] Step S203: Input each of the above first feature vectors into the credibility model to obtain the credibility of each of the above users. The credibility model is a credibility analysis model obtained by training through reinforcement learning or genetic algorithm.

[0180] Specifically, the credibility analysis model, trained through reinforcement learning or genetic algorithms, can update the user credibility in real time within the range of 0-1 by updating the first feature vector.

[0181] Step S204: Filter out the comments from the users whose credibility is lower than the predetermined credibility, and obtain the reference comments for the target product.

[0182] Specifically, low-scoring user reviews are automatically filtered out based on a profile credibility threshold, such as those with a credibility score of less than 0.3.

[0183] Step S205: Aggregate the referenced reviews of the target product to generate a review summary of the target product, and generate improvement suggestions for the target product based on the feedback content of the review summary.

[0184] Specifically, the retained, referable reviews are aggregated by weight, for example, using TF-IDF and BERT semantic clustering, to generate review summaries and store them in Elasticsearch. Based on the feedback content of the review summaries, improvement suggestions for the target product can be generated. For example, based on feedback about short battery life, improvement suggestions for increasing battery capacity can be generated.

[0185] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. They can be implemented using computer-executable program code, and thus can be stored in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those described herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0186] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0187] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0188] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0189] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0190] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0191] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0192] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0193] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0194] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0195] As can be seen from the above description, the embodiments of this application achieve the following technical effects:

[0196] 1) In the product improvement suggestion generation method of this application, the credibility of users is analyzed by using the user characteristic information of the target product to filter out comments from users whose credibility is lower than a predetermined credibility. Improvement suggestions for the target product can then be generated based on the remaining referable comments, ensuring that the referable comments can truly reflect the quality of the target product. This generates improvement suggestions that can improve user satisfaction, allowing manufacturers to improve the target product based on the improvement suggestions to increase user satisfaction. This solves the problem in the prior art where online comments are distorted due to user subjective factors, making the improvement suggestions obtained from comment analysis unable to improve user satisfaction.

[0197] 2) In the product improvement suggestion generation device of this application, the credibility of users is analyzed by using the user characteristic information of the target product to filter out comments from users whose credibility is lower than a predetermined credibility. Improvement suggestions for the target product can then be generated based on the remaining referable comments, ensuring that the referable comments can truly reflect the quality of the target product. This generates improvement suggestions that can improve user satisfaction, allowing manufacturers to improve the target product based on the improvement suggestions to increase user satisfaction. This solves the problem in the prior art where online comments are distorted due to user subjective factors, making the improvement suggestions obtained from comment analysis unable to improve user satisfaction.

[0198] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for generating product improvement suggestions, characterized in that, include: Obtain the user's characteristic information for the target product to obtain the first characteristic information. The characteristic information includes one or more of the user's historical behavior, historical comments, and social influence on the consumption platform. The historical behavior includes the frequency of consumption, and the social influence includes the number of followers and the interaction rate. The first feature information of each user is used to form the first feature vector of each user; Each of the first feature vectors is input into the credibility model to obtain the credibility of each user. The credibility model is a credibility analysis model trained by reinforcement learning or genetic algorithm. Filter out user reviews with credibility levels below a predetermined level to obtain referable reviews for the target product; The referenced reviews of the target product are aggregated to generate a review summary of the target product, and improvement suggestions for the target product are generated based on the problem feedback content of the review summary.

2. The method according to claim 1, characterized in that, Before inputting the feature vectors of each user into the credibility model to obtain the credibility of each user, the method further includes: The first acquisition step involves acquiring the feature information of other users to obtain the second feature information, wherein the other users are users of other products; The combination step involves combining the second feature information of each of the other users to form the second feature vector of each of the other users; The analysis step involves inputting each of the second feature vectors into the initial credibility model to obtain the credibility of each of the other users. The filtering step filters out comments from other users whose credibility is lower than the predetermined credibility, and obtains referenceable comments for the other products; The first generation step involves generating a summary of reviews for the other products based on the aggregated referable reviews for those other products. The second acquisition step involves acquiring the click-through rate and report rate of the review summaries of the other products. The click-through rate is the percentage of users who clicked to view the review summaries among the users who browsed the other products, and the report rate is the percentage of users who reported the review summaries as untrue among the users who clicked to view the review summaries. The calculation steps are as follows: calculate the product of the click-through rate and the corresponding weight to obtain a positive reward; calculate the product of the report rate and the corresponding weight to obtain a negative reward; calculate the difference between the positive reward and the negative reward to obtain the reward value of the comment summary. Each time the parameters of the initial credibility model are adjusted, the first acquisition step, the combination step, the analysis step, the filtering step, the first generation step, the second acquisition step, and the calculation step are repeated once in sequence until the maximum reward value of the comment summary is obtained. The initial credibility model corresponding to the maximum reward value is then determined as the credibility model.

3. The method according to claim 2, characterized in that, Before inputting the feature vectors of each user into the credibility model to obtain the credibility of each user, the method further includes: The second generation step involves randomly generating multiple model parameter sets to obtain an initial population. The model parameter sets include all the parameters to be optimized in the initial credibility model. The steps are defined as follows: the reward value of the comment summary corresponding to the model parameter set is determined as the fitness of the individual in the corresponding initial population; The adjustment steps involve identifying individuals with fitness greater than a predetermined fitness as high-quality individuals, performing crossover operations on multiple high-quality individuals to obtain multiple offspring individuals, and randomly adjusting the parameters of at least one offspring individual to obtain at least one mutated individual. The replacement step involves replacing the individuals with fitness lower than the predetermined fitness with the mutated individuals and the unmutated offspring individuals to obtain an updated population. The second generation step, the determination step, the adjustment step, and the replacement step are repeated a predetermined number of times to obtain the latest population. The model parameter set corresponding to the individual with the highest fitness in the latest population is determined as the optimal model parameter set; The initial credibility model corresponding to the optimal model parameter set is determined as the credibility model.

4. The method according to claim 1, characterized in that, Inputting each of the first feature vectors into the credibility model yields the credibility of each user, including: The sparse autoencoder in the credibility model is used to reduce the dimensionality of each first feature vector to obtain multiple first compressed feature vectors. Each of the first compressed feature vectors is input into the GBDT-Transformer fusion model in the credibility model to obtain the credibility of each user.

5. The method according to claim 1, characterized in that, The referable reviews of the target product are aggregated to generate a review summary of the target product, including: The weight of each referenced comment is calculated based on all the referenced comments and the corresponding user's credibility. In the case where multiple referential comments have the same semantic substance, the referential comment with the highest weight is retained; The remaining referable comments are arranged in descending order of their weights to obtain the comment summary.

6. The method according to claim 5, characterized in that, The weight of each referenced comment is calculated based on all the referenced comments and the corresponding user's credibility, including: Calculate the sum of the credibility scores of all the users corresponding to the referenced comments to obtain the cumulative credibility score; Calculate the ratio of the credibility of the user corresponding to each of the reference comments to the cumulative credibility to obtain the credibility weight of each of the reference comments; Each of the referenced comments is converted into a feature vector, resulting in multiple comment feature vectors; Calculate the average of the multiple comment feature vectors to obtain the average feature vector; Calculate the cosine similarity between each of the comment feature vectors and the average feature vector to obtain the similarity weight of each of the reference comments; The weight of each reference comment is obtained by multiplying its credibility weight and its similarity weight.

7. The method according to any one of claims 1 to 6, characterized in that, Before filtering out reviews from users whose credibility is lower than a predetermined credibility level to obtain referable reviews for the target product, the method further includes: The credibility of the user for the target product is updated once at a predetermined interval.

8. An apparatus for generating product improvement suggestions, characterized in that, include: The first acquisition unit is used to acquire the user's feature information of the target product and obtain the first feature information. The feature information includes one or more of the user's historical behavior, historical comments and social influence on the consumption platform. The historical behavior includes the frequency of consumption and the social influence includes the number of fans and the interaction rate. The first combination unit is used to combine the first feature information of each user into a first feature vector of each user. The first analysis unit is used to input each of the first feature vectors into the credibility model to obtain the credibility of each user. The credibility model is a credibility analysis model trained by reinforcement learning or genetic algorithm. The first filtering unit is used to filter out the user's comments whose credibility is lower than a predetermined credibility, and obtain the referenceable comments of the target product; The first generation unit is used to aggregate the referable reviews of the target product to generate a review summary of the target product, and generate improvement suggestions for the target product based on the problem feedback content of the review summary.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform the method for generating product improvement suggestions as described in any one of claims 1 to 7.

10. A comment analysis system, characterized in that, include: One or more processors, a memory, and one or more programs, wherein the review analysis system is applied to a consumer platform, the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include a method for generating product improvement suggestions according to any one of claims 1 to 7.

Images