Automobile personalized recommendation method based on comment text and score fusion

Abstract

The embodiment of the invention provides an automobile personalized recommendation method based on comment text and score fusion, and belongs to the technical field of data processing, and the method specifically comprises the steps: obtaining a user star-level score, comment text data and user feature data; performing emotion polarity analysis to obtain an emotion score vector; obtaining a deterministic score vector; constructing an undirected network and dividing user groups with the same star-level score; obtaining a first comprehensive text emotion score vector, a first comprehensive deterministic score vector and a first comprehensive length; calculating an aggregation weight of each community and obtaining a second comprehensive text sentiment score vector and a second deterministic score vector under the same star-level score according to the aggregation weight; calculating a discount coefficient of the star-level score, and obtaining a comprehensive score under each attribute; according to the personalized attribute weights and the comprehensive scores of the users, comprehensive scores of all the alternative automobiles are obtained through weighted average, and finally the optimal automobile is recommended. Through the scheme of the invention, the recommendation accuracy and comprehensiveness are improved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of data processing technology, and in particular to a personalized automobile recommendation method based on the fusion of review text and rating. Background Art

[0002] At present, with the rapid development of social economy and new energy vehicles, the number of car drivers is gradually increasing. As a large-value consumer product, user reviews of cars have important reference value for other consumers' car purchase decisions. Therefore, the study of sentiment polarity, certainty analysis and user star rating of car user review text is of great significance.

[0003] As for user ratings, ratings are intuitive and easy to understand information, which also has important reference value for consumers' car purchase decisions. However, the limited rating differentiation does not allow users to distinguish products better. Since the ratings usually have only five or several levels, the subtle differences when users express their satisfaction or dissatisfaction with the products may not be fully reflected. For example, users may have similar satisfaction with two different products, but due to the restrictions on the rating levels, they may give the same rating. In addition, users have limited time and energy, and their understanding online and offline will be limited, which may cause some better models and reviews to not enter the eyes of users and be ignored.

[0004] It can be seen that there is an urgent need for a personalized car recommendation method based on the fusion of review text and ratings with high recommendation accuracy and comprehensiveness. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a personalized car recommendation method based on the fusion of review text and rating, which at least partially solves the problem of poor recommendation accuracy and comprehensiveness in the prior art.

[0006] The embodiment of the present invention provides a personalized automobile recommendation method based on the fusion of review text and rating, comprising: Step 1, obtaining user star ratings, comment text data and user feature data from multiple consulting platforms, wherein the user star ratings include attributes of vehicle space, handling, driving experience, comfort, fuel consumption, appearance, interior, cost performance and configuration, and the user feature data includes user authentication information, platform recommendation level for comments, and text comment interaction information; Step 2: Perform sentiment polarity analysis on the preprocessed comment text data based on the structBERT large language model to obtain a sentiment score vector corresponding to the comment text data, wherein the sentiment score vector includes three dimensions: positive, neutral, and negative. Step 3, constructing a review text certainty dictionary through manual annotation, and obtaining a certainty score vector corresponding to the review text data based on word frequency statistics and the certainty dictionary, wherein the certainty score vector includes five dimensions: the total certainty score of the certainty words, the number of certainty words, the maximum certainty score, the minimum certainty score, and the mean certainty score; Step 4: Based on the sentiment score vector, certainty score vector and text length of the text review data, the similarity between the review text data of different users is calculated, and an undirected network with users as nodes and similarity as node relationships is constructed accordingly. The Louvain algorithm is used to divide the user groups with the same star rating into multiple communities. Step 5, calculating the aggregation weight of each user in the community according to the user feature data, and based on this, obtaining the first comprehensive text sentiment score vector, the first comprehensive certainty score vector and the first comprehensive length of each community through weighted average operator aggregation; Step 6, according to the first comprehensive text sentiment score vector, the first comprehensive certainty score vector, the standard deviation of the first comprehensive length and the number of community users, calculate the aggregation weight of each community and obtain the second comprehensive text sentiment score vector and the second certainty score vector under the same star rating through weighted average operator aggregation; Step 7, based on the distance between the second comprehensive text sentiment score vector and the star rating, combined with the second certainty score vector, calculate the discount coefficient of the star rating t, calculate the aggregation weight of each star rating based on the number of users who rated star rating t, and obtain the comprehensive score of the car under each attribute through weighted average operator aggregation based on the discount coefficient and the aggregation weight of each star rating; Step 8: Based on the personalized attribute weights and comprehensive scores set by the user, the comprehensive scores of all candidate cars are obtained through weighted average operator aggregation, and then the ranking of all cars is obtained to recommend the best car.

[0007] According to a specific implementation of the embodiment of the present invention, step 2 specifically includes: Use the structBERT model to obtain comment text data The sentiment score vector , which satisfies: in, , Indicates star rating Corresponding review text data, star rating Indicates car property Next User The star rating given is t, t=1,2,3,4,5.

[0008] According to a specific implementation of the embodiment of the present invention, step 3 specifically includes: Step 3.1: Construct each attribute through manual annotation A deterministic dictionary, wherein the deterministic dictionary includes a word set = and its corresponding set of certainty score vectors ,in Representation attributes No. A definite word, Words that express certainty The certainty score vector of ; Step 3.2: Obtain review text data based on word frequency statistics and deterministic dictionaries The corresponding certainty score vector ; in, Represents comment text data The total certainty score of all certainty words in Indicates the number of certain words, represents the maximum certainty score, represents the minimum certainty score, represents the mean certainty score.

[0009] According to a specific implementation of the embodiment of the present invention, step 4 specifically includes: Step 4.1, calculate the comment text data of different users based on the sentiment score vector, certainty score vector and text length of the text comment data and The sentiment score vector similarity, certainty score vector similarity and text length similarity between them, where the expression of the sentiment score vector similarity is: ; in, Represents comment text data The sentiment score vector of Represents comment text data Sentiment score vector of ; The expression of the deterministic score vector similarity is: ; in, Represents comment text data The sentiment score vector of Represents comment text data Sentiment score vector of ; The expression of the text length similarity is: in, Represents comment text data The length of the text, Represents comment text data Length of text; Step 4.2, calculate the threshold based on the sentiment score vector similarity, certainty score vector similarity and text length similarity matrix ; Step 4.3: construct Build an undirected network with users as nodes and the similarity between comment texts as node relationships. Its adjacency matrix satisfy: Step 4.4: Use the Louvain community discovery algorithm to divide user groups with the same star rating ; Get G communities ; , ; With the community The corresponding text set satisfy: ; Among them, the text The corresponding sentiment score vector is , the certainty score vector is and length is .

[0010] According to a specific implementation of the embodiment of the present invention, step 5 specifically includes: Step 5.1: Calculate the user's rating based on the user authentication information, the platform's recommendation level for the user's comments, and the text comment interaction data. Aggregation weight ; in, For text The recommended level value of For the platform The authentication value of For text Harvest interaction values ​​from other users, where the interaction values ​​include text Gain likes, views and secondary comments from other users; Step 5.2, based on aggregate weight Get Community The first comprehensive text sentiment score vector , the first comprehensive certainty score vector and the first comprehensive length .

[0011] According to a specific implementation of the embodiment of the present invention, step 6 specifically includes: Step 6.1, calculate each community Aggregation weight ; Among them, the adjustment parameters , Based on The weight of the number of community users, satisfying , It is a comprehensive standard deviation constructed based on the standard deviation of the sentiment score vector, the standard deviation of the certainty score vector, and the length standard deviation of all texts in the community, and is normalized to meet the following requirements: in: Represent the standard deviation of the first comprehensive text sentiment score vector, the first comprehensive certainty score vector and the first comprehensive length of all texts in the community respectively; Step 6.2, calculate the second comprehensive text sentiment score vector under the same star rating based on the aggregate weight of each community and the second certainty score vector : .

[0012] According to a specific implementation of the embodiment of the present invention, step 7 specifically includes: Step 7.1, Rating by Stars The corresponding second comprehensive text sentiment score vector distance, calculate the star rating Discount factor : in, is the mean certainty score, Represents the normalized value of the sentiment score vector ; Step 7.2, Rating by Stars The corresponding number of users , calculate the aggregation weight of the star rating ; = ; Step 7.3, calculate the car's Properties Overall rating .

[0013] According to a specific implementation of the embodiment of the present invention, step 8 specifically includes: Step 8.1: Set properties according to user's personalized needs Weight ; Step 8.2: Get candidate cars based on weighted average operator aggregation Overall rating ; ; Step 8.3, based on the comprehensive score Sort all the candidate cars and recommend the best car based on the ranking of all the cars.

[0014] The personalized car recommendation scheme based on the fusion of comment text and rating in the embodiment of the present invention includes: step 1, obtaining user star ratings, comment text data and user feature data in multiple consulting platforms, wherein the user star ratings include attributes of car space, control, driving experience, comfort, fuel consumption, appearance, interior, cost performance and configuration, and the user feature data includes user authentication information, platform recommendation level for comments, and text comment interaction information; step 2, performing sentiment polarity analysis on the pre-processed comment text data based on the structBERT large language model to obtain the sentiment score corresponding to the comment text data vector, wherein the sentiment score vector includes three dimensions: positive, neutral and negative; step 3, constructing a comment text deterministic dictionary through manual annotation, and obtaining the certainty score vector corresponding to the comment text data based on word frequency statistics and the deterministic dictionary, wherein the certainty score vector includes five dimensions: the total certainty score of the deterministic words, the number of certainty words, the maximum certainty score, the minimum certainty score and the mean certainty score; step 4, based on the sentiment score vector, the certainty score vector and the text length of the text comment data, calculating the similarity between the comment text data of different users, and constructing a user-based node based on this. , similarity as the node relationship of the undirected network, using the Louvain algorithm to divide the user groups with the same star rating to obtain multiple communities; step 5, according to the user feature data, the aggregation weight of each user in the community is calculated, and the first comprehensive text sentiment score vector, the first comprehensive certainty score vector and the first comprehensive length of each community are obtained by weighted average operator aggregation; step 6, according to the standard deviation of the first comprehensive text sentiment score vector, the first comprehensive certainty score vector and the first comprehensive length and the number of community users, the aggregation weight of each community is calculated, and the second comprehensive length under the same star rating is obtained by weighted average operator aggregation. The text sentiment score vector and the second deterministic score vector are combined; step 7, based on the distance between the second comprehensive text sentiment score vector and the star rating, the discount coefficient of the star rating t is calculated in combination with the second deterministic score vector, the aggregation weight of each star rating is calculated based on the number of users of the star rating t, and the comprehensive score of the car under each attribute is obtained by weighted average operator aggregation based on the discount coefficient and the aggregation weight of each star rating; step 8, based on the personalized attribute weights and comprehensive scores set by the user, the comprehensive scores of all candidate cars are obtained by weighted average operator aggregation, and then the ranking of all cars is obtained, and the best car is recommended.

[0015] The beneficial effects of the embodiments of the present invention are as follows: through the scheme of the present invention, based on user characteristic data such as user star ratings of multiple attributes of cars in multiple consulting platforms, comment text data and user authentication information, platform recommendation levels for comments, text comment interaction information, etc.; comment text data is extracted and preprocessed; according to natural language processing technology, sentiment polarity analysis of comments is performed based on the structBERT large language model to obtain comment sentiment score vectors; a comment text deterministic dictionary is constructed based on the word frequency statistics of the comment text and manual annotations to obtain a deterministic score vector of the preprocessed comment text; a similarity matrix is ​​calculated based on the length of the comment text sentiment score vector and the deterministic score vector, a graph network is constructed with users as nodes and similarities as node relationships, and community discovery algorithm division is performed based on the similarity threshold; based on user authentication information and platform comments on users, a community discovery algorithm is used to divide the community into different categories. The recommendation level of the text, comment interaction information and other user feature data are used to obtain the comprehensive text sentiment score vector, certainty score vector and length of each community under the same star rating; according to the standard deviation of the sentiment score, certainty score and length of the comment text of each community user, the aggregation weight of each community is calculated in combination with the number of community users, and the comprehensive text sentiment score and certainty score under the same star rating are obtained through weighted average operator aggregation; according to the distance between the comment text sentiment score and the star rating, combined with the certainty score, the discount coefficient of the star rating is obtained, and the aggregation weight based on the number of users with star ratings is used to obtain the comprehensive score under each attribute of the car model; based on the personalized attribute weight, the comprehensive score of all candidate cars is obtained through weighted average operator aggregation, and then the ranking of all cars is obtained, and the optimal car is recommended, providing consumers with more comprehensive car purchase reference information. Compared with the prior art, the present invention provides users with convenient and efficient comment text analysis and car recommendation from multiple aspects of user ratings and comment texts, thereby improving the comprehensiveness of car recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 A flowchart of a method for personalized car recommendation based on the fusion of review text and ratings provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0019] The following describes the embodiments of the present invention through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0020] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present invention, it should be understood by those skilled in the art that an aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects described herein can be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein can be used to implement this device and / or practice this method.

[0021] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention. The drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0022] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects described may be practiced without these specific details.

[0023] In past studies, many researchers have conducted in-depth research on sentiment analysis of user review texts, mainly using technologies such as text mining and natural language processing, and classifying reviews by constructing sentiment dictionaries and machine learning. These review text studies mainly focus on sentiment classification, that is, judging whether the review is positive, negative, or neutral. Although this method can capture the user's emotional tendencies, it often ignores the degree of certainty of the review. For example, one user may say "the appearance, interior, and handling performance of this car are great", while another user may say "this car is not bad". Both express positive emotions, but the former's review text expression has a higher degree of certainty.

[0024] In addition, user ratings, as an intuitive and easy-to-understand information, are also of great reference value for consumers' car purchase decisions. However, the limited rating differentiation does not allow users to better distinguish products. Since the ratings usually have only five or several levels, this may result in the subtle differences when users express satisfaction or dissatisfaction with the products not being fully reflected. For example, users may have similar satisfaction with two different products, but due to the restrictions on the rating levels, they may give the same rating. In addition, users have limited time and energy, and their understanding online and offline will be limited, which may result in some better models and reviews not being seen by users and being ignored.

[0025] The embodiment of the present invention provides a personalized automobile recommendation method based on the fusion of review text and rating, and the method can be applied to the automobile consultation recommendation process in the Internet scenario.

[0026] See also Figure 1 , is a flow chart of a personalized car recommendation method based on the fusion of review text and rating provided by an embodiment of the present invention. Figure 1 As shown, the method mainly comprises the following steps: Step 1, obtaining user star ratings, comment text data and user feature data from multiple consulting platforms, wherein the user star ratings include attributes of vehicle space, handling, driving experience, comfort, fuel consumption, appearance, interior, cost performance and configuration, and the user feature data includes user authentication information, platform recommendation level for comments (including full level essence, essence, recommended, ordinary), and text comment interaction information (including likes, views, and secondary comments); Step 2: Performing preprocessing such as deduplication and word segmentation on the car user review text data based on the structBERT large language model, and performing sentiment polarity analysis on the preprocessed review text data to obtain a sentiment score vector corresponding to the review text data, wherein the sentiment score vector includes three dimensions: positive, neutral, and negative; Step 3, constructing a review text certainty dictionary through manual annotation, and obtaining a certainty score vector corresponding to the review text data based on word frequency statistics and the certainty dictionary, wherein the certainty score vector includes five dimensions: the total certainty score of the certainty words, the number of certainty words, the maximum certainty score, the minimum certainty score, and the mean certainty score; Step 4: Based on the sentiment score vector, certainty score vector and text length of the text review data, the similarity between the review text data of different users is calculated, and an undirected network with users as nodes and similarity as node relationships is constructed accordingly. The Louvain algorithm is used to divide the user groups with the same star rating into multiple communities. Step 5, calculating the aggregation weight of each user in the community according to the user feature data, and based on this, obtaining the first comprehensive text sentiment score vector, the first comprehensive certainty score vector and the first comprehensive length of each community through weighted average operator aggregation; Step 6, according to the first comprehensive text sentiment score vector, the first comprehensive certainty score vector, the standard deviation of the first comprehensive length and the number of community users, calculate the aggregation weight of each community and obtain the second comprehensive text sentiment score vector and the second certainty score vector under the same star rating through weighted average operator aggregation; Step 7, based on the distance between the second comprehensive text sentiment score vector and the star rating, combined with the second certainty score vector, calculate the discount coefficient of the star rating t, calculate the aggregation weight of each star rating based on the number of users who rated star rating t, and obtain the comprehensive score of the car under each attribute through weighted average operator aggregation based on the discount coefficient and the aggregation weight of each star rating; Step 8: Based on the personalized attribute weights and comprehensive scores set by the user, the comprehensive scores of all candidate cars are obtained through weighted average operator aggregation, and then the ranking of all cars is obtained to recommend the best car.

[0027] Specifically, step 1 includes: The embodiment of the present invention obtains data of multiple target automobile consulting platforms through Python tools, such as user star ratings of multiple attributes such as automobile space, control, driving experience, comfort, fuel consumption, appearance, interior, cost performance, configuration, etc. in the reputation module of the automobile rating reputation platform, comment text data, and user authentication information (including authentication, unauthentication), platform recommendation level for comments (including full level essence, essence, recommendation, ordinary), text comment interaction data (including likes, browsing, secondary comments) and other user feature data; and classifies them according to car models and stores them in the local database, among which the platform websites of the target automobile consulting platforms are selected as: Autohome, Dongchedi, Aika Auto and Pacific Auto, etc.; Among them, car attributes include space, control, driving experience, comfort, fuel consumption, appearance, interior, cost performance, configuration, etc.; User star ratings are usually t=1, 2, 3, 4, 5 points.

[0028] As shown in Table 1 below, the (Passat) Attributes (Space) Star rating is 5 (t=5) users The data example has the comment text .

[0029] Table 1

[0030] Specifically, step 2 includes: First, we can use the structBERT model to obtain the comment text The sentiment score vector , and satisfies Table 2 provides the car images obtained based on structBERT. (Passat) Attributes (Space) Star rating out of 5 review text Example of sentiment score vector for .

[0031] Table 2

[0032] Specifically, step 3 includes: Calculate the review text based on the review text deterministic dictionary The certainty score vector As shown below, the certainty word score range is [0,9], and the higher the score, the higher the degree of certainty: ; First, Table 3 provides the car-oriented attributes (spatial) Partial Deterministic Dictionary.

[0033] Table 3

[0034] Compute review text based on a deterministic dictionary The certainty score vector Table 4 provides the car property (space) Star rating t=5 review text The certainty score vector of .

[0035] Table 4

[0036] Specifically, step 4 includes: Construct an undirected network with users as nodes and the similarity between review texts as node relationships, and use the Louvain community discovery algorithm to find the community division results of user groups under star ratings; First, calculate the car property Thresholds for sentiment score vector similarity, certainty score vector similarity, and text length similarity , which are 0.93, 0.91 and 0.86 respectively.

[0037] The undirected network constructed with users as nodes and the similarity between comment texts as node relationships has an adjacency matrix ,satisfy: Table 5: Automobile property 2000 users with star rating t=5 For example, the Louvain algorithm obtains the results of four community divisions of the user group.

[0038] Table 5

[0039] After deleting the outlier nodes, the community division results are obtained.

[0040] Specifically, step 5 includes: According to the user's aggregation weight, the sentiment score vector, certainty score vector and length of all user texts in each community are aggregated.

[0041] First, assign values ​​to user authentication information, the platform's recommendation level for user comment text, and comment interaction data. The standards are as follows: User authentication value assignment: 1 for authenticated and 0 for unauthenticated.

[0042] The recommended level assignment for comment text is: 1 for full level essence, 0.7 for essence, 0.5 for recommended, and 0.3 for ordinary.

[0043] Interaction data assignment: According to the comprehensive sum of the number of likes, views, secondary comments, etc., the following five situations are assigned values: 1, 0.8, 0.6, 0.4, 0.2: sum>20000, 20000>sum>10000, 10000>sum>5000, 5000>sum>500.

[0044] Counting users Aggregation weight : in, The value settings are as follows Then, based on the above weights, get the community The comprehensive sentiment score vector of the text , comprehensive certainty score vector and the overall length The following car property The results for the g=1 community with a star rating of t=5 are as follows: The comprehensive text sentiment score vector of the community (g=1) =[0.55,0.16,0.14]; The comprehensive text certainty score vector of the community (g=1) ; The total length of text in the community (g=1) .

[0045] Specifically, step 6 includes: Calculate the comprehensive sentiment score vector and certainty score vector of all community user texts under the same star rating.

[0046] First, calculate each community Aggregation weight ( ). property Take the aggregation weights of the four communities with star rating t=5 in (space) as an example, and set ,get , , , .

[0047] Then, calculate the car property The comprehensive text sentiment score vector and certainty score vector of star rating t=5 in (space).

[0048] Specifically, step 7 includes: First, based on the rating (Indicates car property Star rating t) and its corresponding text sentiment score vector distance, calculate the star rating Discount factor . property For example, get the discount factor of star rating t (t=1,2,3,4,5,): Then, based on the star rating The corresponding number of users , calculate the aggregation weight of the star rating . property For example, get the star rating t (t=1,2,3,4,5,) aggregation weight: , , , , ; Then calculate the car property Overall Rating ; Specifically, step 8 includes: Assuming the user is a personalized appearance preference, set the property Weight As shown in Table 6.

[0049] Table 6

[0050] Get the candidate cars by weighted average operator aggregation Overall rating The ranking results are shown in Table 7. The best one is .

[0051] Table 7

[0052] The personalized car recommendation method based on the fusion of comment text and rating provided in this embodiment is based on user star ratings of multiple attributes of cars in multiple consulting platforms, comment text data and user authentication information, platform recommendation level for comments, text comment interaction information and other user feature data; extract and preprocess the comment text data; according to natural language processing technology, based on the structBERT large language model, perform sentiment polarity analysis on the comments to obtain the comment sentiment score vector; construct a comment text deterministic dictionary based on the word frequency statistics of the comment text and manual annotation, and obtain the deterministic score vector of the preprocessed comment text; calculate the similarity matrix based on the length of the comment text sentiment score vector and the deterministic score vector, construct a graph network with users as nodes and similarity as node relationships, and divide the community discovery algorithm based on the similarity threshold; according to user authentication information, platform The user characteristic data such as the recommendation level and comment interaction information of the user comment text are used to obtain the comprehensive text sentiment score vector, certainty score vector and length of each community under the same star rating; the aggregation weight of each community is calculated based on the standard deviation of the sentiment score, certainty score and length of the user comment text of each community, combined with the number of community users, and the comprehensive text sentiment score and certainty score under the same star rating are obtained through weighted average operator aggregation; the discount coefficient of the star rating is obtained based on the distance between the comment text sentiment score and the star rating, combined with the certainty score, and the comprehensive score of each attribute under the car model is obtained based on the aggregation weight of the number of users with star ratings; based on the personalized attribute weight, the comprehensive score of all candidate cars is obtained through weighted average operator aggregation, and then the ranking of all cars is obtained, and the optimal car is recommended, providing consumers with more comprehensive car purchase reference information. Compared with the prior art, the present invention provides users with convenient and efficient comment text analysis and car recommendation from multiple aspects of user ratings and comment texts, thereby improving the comprehensiveness of car recommendations.

[0053] It should be understood that various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof.

[0054] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A personalized car recommendation method based on the fusion of review text and rating, characterized in that: include: Step 1, obtaining user star ratings, comment text data and user feature data from multiple consulting platforms, wherein the user star ratings include attributes of vehicle space, handling, driving experience, comfort, fuel consumption, appearance, interior, cost performance and configuration, and the user feature data includes user authentication information, platform recommendation level for comments, and text comment interaction information; Step 2: Perform sentiment polarity analysis on the preprocessed comment text data based on the structBERT large language model to obtain a sentiment score vector corresponding to the comment text data, wherein the sentiment score vector includes three dimensions: positive, neutral, and negative. Step 3, constructing a review text certainty dictionary through manual annotation, and obtaining a certainty score vector corresponding to the review text data based on word frequency statistics and the certainty dictionary, wherein the certainty score vector includes five dimensions: the total certainty score of the certainty words, the number of certainty words, the maximum certainty score, the minimum certainty score, and the mean certainty score; Step 4: Based on the sentiment score vector, certainty score vector and text length of the text review data, the similarity between the review text data of different users is calculated, and an undirected network with users as nodes and similarity as node relationships is constructed accordingly. The Louvain algorithm is used to divide the user groups with the same star rating into multiple communities. Step 5, calculating the aggregation weight of each user in the community according to the user feature data, and based on this, obtaining the first comprehensive text sentiment score vector, the first comprehensive certainty score vector and the first comprehensive length of each community through weighted average operator aggregation; Step 6, according to the first comprehensive text sentiment score vector, the first comprehensive certainty score vector, the standard deviation of the first comprehensive length and the number of community users, calculate the aggregation weight of each community and obtain the second comprehensive text sentiment score vector and the second certainty score vector under the same star rating through weighted average operator aggregation; Step 7, based on the distance between the second comprehensive text sentiment score vector and the star rating, combined with the second certainty score vector, calculate the discount coefficient of the star rating t, calculate the aggregation weight of each star rating based on the number of users who rated star rating t, and obtain the comprehensive score of the car under each attribute through weighted average operator aggregation based on the discount coefficient and the aggregation weight of each star rating; Step 8: Based on the personalized attribute weights and comprehensive scores set by the user, the comprehensive scores of all candidate cars are obtained through weighted average operator aggregation, and then the ranking of all cars is obtained to recommend the best car.

2. The method according to claim 1, characterized in that , the step 2 specifically includes: Use the structBERT model to obtain comment text data The sentiment score vector , which satisfies: in, , Indicates star rating Corresponding review text data, star rating Indicates car property Next User The star rating given is t, t=1,2,3,4,5.

3. The method according to claim 2, characterized in that , the step 3 specifically includes: Step 3.1: Construct each attribute through manual annotation A deterministic dictionary, wherein the deterministic dictionary includes a word set = and its corresponding set of certainty score vectors ,in Representation attributes No. A definite word, Words that express certainty The certainty score vector of ; Step 3.2: Obtain review text data based on word frequency statistics and deterministic dictionaries The corresponding certainty score vector ; in, Represents comment text data The total certainty score of all certainty words in Indicates the number of certain words, represents the maximum certainty score, represents the minimum certainty score, represents the mean certainty score.

4. The method according to claim 3, characterized in that , the step 4 specifically includes: Step 4.1, calculate the comment text data of different users based on the sentiment score vector, certainty score vector and text length of the text comment data and The sentiment score vector similarity, certainty score vector similarity and text length similarity between them, where the expression of the sentiment score vector similarity is: ； in, Represents comment text data The sentiment score vector of Represents comment text data Sentiment score vector of ; The expression of the deterministic score vector similarity is: in, Represents comment text data The sentiment score vector of Represents comment text data Sentiment score vector of ; The expression of the text length similarity is: in, Represents comment text data The length of the text, Represents comment text data Length of text; Step 4.2, calculate the threshold based on the sentiment score vector similarity, certainty score vector similarity and text length similarity matrix ; Step 4.3: construct Build an undirected network with users as nodes and the similarity between comment texts as node relationships. Its adjacency matrix satisfy: Step 4.4: Use the Louvain community discovery algorithm to divide user groups with the same star rating ; Get G communities ; , ； Community The corresponding text set satisfy: ； Among them, the text The corresponding sentiment score vector is , certainty score vector and the length is .

5. The method according to claim 4, characterized in that , the step 5 specifically includes: Step 5.1: Calculate the user's rating based on the user authentication information, the platform's recommendation level for the user's comments, and the text comment interaction data. Aggregation weight ; in, For Text The recommended level value of For the platform The authentication value of For Text Harvest interaction values ​​from other users, where the interaction values ​​include text Gain likes, views and secondary comments from other users; Step 5.2, based on aggregate weight Get Community The first comprehensive text sentiment score vector , the first comprehensive certainty score vector and the first comprehensive length .

6. The method according to claim 5, characterized in that , the step 6 specifically includes: Step 6.1, calculate each community Aggregation weight ; Among them, the adjustment parameters , Based on The weight of the number of community users, satisfying , It is a comprehensive standard deviation constructed based on the standard deviation of the sentiment score vector, the standard deviation of the certainty score vector, and the length standard deviation of all texts in the community, and is normalized to meet the following requirements: in: , , Represent the standard deviation of the first comprehensive text sentiment score vector, the first comprehensive certainty score vector and the first comprehensive length of all texts in the community respectively; Step 6.2, calculate the second comprehensive text sentiment score vector under the same star rating based on the aggregate weight of each community and the second certainty score vector : 。 7. The method according to claim 6, characterized in that , the step 7 specifically includes: Step 7.1, Rating by Stars The corresponding second comprehensive text sentiment score vector distance, calculate the star rating Discount factor : in, is the mean certainty score, Represents the normalized value of the sentiment score vector ； Step 7.2, Rating by Stars The corresponding number of users , calculate the aggregation weight of the star rating ; ； Step 7.3, calculate the car's Properties Overall rating 。 8. The method according to claim 7, characterized in that , the step 8 specifically includes: Step 8.1: Set properties according to user's personalized needs Weight ; Step 8.2: Get candidate cars based on weighted average operator aggregation Overall rating ; ； Step 8.3, based on the comprehensive score Sort all the candidate cars and recommend the best car based on the ranking of all the cars.

Images