Method, system and terminal device for modifying a score of an evaluation text

Abstract

The embodiment of the application provides a kind of to the score modification method, system and terminal equipment of evaluation text, belong to artificial intelligence technical field.The method includes: obtaining historical consumption information and initial evaluation text and initial score information under historical consumption information;Text sentiment consistency analysis is carried out to initial evaluation text to obtain target analysis result;According to target analysis result, initial score information is determined in combination with target data state;The information correlation degree between historical consumption information and initial evaluation text is calculated;According to target data state and information correlation degree, initial evaluation text is deleted to obtain target evaluation text, and associated score information is obtained from initial score information;Target evaluation text is carried out data clustering to obtain target clustering result, and according to target clustering result, associated score information is modified to obtain target score information.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, system, and terminal device for modifying the rating of evaluation text. Background Technology

[0002] In today's digital commerce environment, recommending stores to users can utilize comprehensive store evaluation scores to filter out stores that meet their needs and preferences, significantly improving shopping efficiency and experience. However, with current technology, the comprehensive evaluation of stores often involves simply weighting and averaging certain data. This results in evaluations that fail to accurately reflect the actual situation of each store, leading to inaccurate overall product evaluation scores. Consequently, this negatively impacts subsequent store recommendations and rankings, ultimately diminishing the user's shopping experience. Summary of the Invention

[0003] The main objective of this invention is to provide a method, system, and terminal device for modifying ratings of evaluation texts, aiming to solve the problem in related technologies where the comprehensive evaluation value of a product is not accurate enough, which can have a serious negative impact on subsequent store recommendations and store rankings, thereby greatly reducing the user's consumption experience.

[0004] In a first aspect, embodiments of the present invention provide a method for modifying the rating of evaluation text, comprising:

[0005] Obtain historical consumption information of the target user after making a purchase at the target merchant, and obtain the initial evaluation text of the target merchant under the historical consumption information and the initial rating information corresponding to the initial evaluation text;

[0006] Perform text sentiment consistency analysis on the initial evaluation text to obtain the target analysis results corresponding to the initial evaluation text;

[0007] Based on the target analysis results and the initial scoring information, determine the target data status corresponding to the initial evaluation text;

[0008] Calculate the information correlation degree between the historical consumption information and the initial evaluation text;

[0009] Based on the target data status and the information correlation, the initial evaluation text is deleted to obtain the target evaluation text, and the associated rating information corresponding to the target evaluation text is obtained from the initial rating information;

[0010] The target evaluation text is clustered to obtain target clustering results, and the associated rating information is corrected according to the target clustering results to obtain target rating information corresponding to the target evaluation text.

[0011] Based on the historical consumption information, the target evaluation text, and the target rating information, an artificial intelligence model is used to determine the merchant value assessment result corresponding to the target merchant.

[0012] Secondly, embodiments of the present invention provide a rating modification system for evaluation text, comprising:

[0013] The data acquisition module is used to obtain historical consumption information corresponding to the target user's consumption under the target merchant, and to obtain the initial evaluation text of the target merchant and the initial rating information corresponding to the initial evaluation text under the historical consumption information.

[0014] The sentiment analysis module is used to perform text sentiment consistency analysis on the initial evaluation text to obtain the target analysis result corresponding to the initial evaluation text;

[0015] The status determination module is used to determine the target data status corresponding to the initial evaluation text based on the target analysis results and the initial scoring information.

[0016] The correlation calculation module is used to calculate the information correlation degree between the historical consumption information and the initial evaluation text;

[0017] The data association module is used to delete the initial evaluation text according to the target data status and the information association degree to obtain the target evaluation text, and to obtain the associated rating information corresponding to the target evaluation text from the initial rating information;

[0018] The data correction module is used to perform data clustering on the target evaluation text to obtain target clustering results, and to perform data correction on the associated rating information based on the target clustering results to obtain target rating information corresponding to the target evaluation text;

[0019] The value assessment module is used to determine the merchant value assessment result corresponding to the target merchant based on the historical consumption information, the target evaluation text, and the target rating information using an artificial intelligence model.

[0020] Thirdly, embodiments of the present invention also provide a terminal device, the terminal device including a processor, a memory, a computer program stored in the memory and executable by the processor, and a data bus for implementing communication between the processor and the memory, wherein when the computer program is executed by the processor, it implements the steps of any of the rating modification methods for evaluation text provided in this specification.

[0021] This invention provides a method, system, and terminal device for modifying ratings of evaluation texts. The method includes: obtaining historical consumption information of a target user after making a purchase at a target merchant, obtaining initial evaluation text and initial rating information of the target merchant based on the historical consumption information, performing sentiment consistency analysis on the initial evaluation text to obtain a target analysis result, determining the target data state of the initial evaluation text based on the target analysis result and the initial rating information, and calculating the information correlation degree between the historical consumption information and the initial evaluation text; deleting the initial evaluation text based on the target data state and information correlation degree to obtain the target evaluation text, obtaining the associated rating information of the target evaluation text from the initial rating information, performing data clustering on the target evaluation text to obtain the target clustering result, and correcting the associated rating information based on the target clustering result to obtain the target rating information of the target evaluation text; finally, using an artificial intelligence model to determine the merchant value assessment result of the target merchant based on the historical consumption information, the target evaluation text, and the target rating information. This method performs sentiment consistency analysis on initial evaluation texts, combines this with initial rating information to determine the target data state of the initial evaluation texts, and then deletes initial evaluation texts based on the target data state and information relevance. This effectively removes evaluations that may be abnormal, false, or have low relevance to consumer information, thus filtering out this invalid information and improving the quality of evaluation data. Furthermore, it performs data clustering on the target evaluation texts and corrects the associated rating information based on the clustering results, enabling ratings to more accurately reflect consumers' true experiences. Finally, by combining historical consumption information, target evaluation texts, and target rating information, an artificial intelligence model is used to determine the merchant value assessment results corresponding to the target merchants. This allows for a comprehensive and objective evaluation of merchant value from multiple dimensions, enabling a more accurate measurement of the merchant's operating status and development potential. This allows users to make more informed decisions when choosing merchants based on this accurate information, reducing consumption risks. This also solves the problem in related technologies where insufficient accuracy in comprehensive product evaluation values ​​can severely negatively impact subsequent store recommendations and rankings, significantly reducing the user's consumption experience. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1This is a flowchart illustrating a method for modifying the rating of evaluation text according to an embodiment of the present invention.

[0024] Figure 2 This is a schematic diagram of the module structure of a rating modification system for evaluation text provided in an embodiment of the present invention;

[0025] Figure 3 This is a schematic block diagram of a terminal device provided in an embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0028] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0029] This invention provides a method, system, and terminal device for modifying ratings on evaluation text. The method for modifying ratings on evaluation text can be applied to a terminal device, which can be an electronic device such as a tablet computer, laptop computer, desktop computer, personal digital assistant, or wearable device. The terminal device can be a server or a server cluster.

[0030] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0031] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a method for modifying the rating of evaluation text, as provided in an embodiment of the present invention.

[0032] like Figure 1 As shown, the method for modifying the rating of the evaluation text includes steps S101 to S107.

[0033] Step S101: Obtain the historical consumption information of the target user after making a purchase at the target merchant, and obtain the initial evaluation text of the target merchant and the initial rating information corresponding to the initial evaluation text under the historical consumption information.

[0034] For example, historical consumption information of target users after making purchases at target merchants is obtained from the database, as well as the initial review text of the target users when they rated the target merchants based on this historical consumption information, and the initial rating information of the target users after they rated the target merchants based on the initial review text, obtained from the database or by using web crawling. The target users are multiple users who made purchases at the target merchants.

[0035] Step S102: Perform text sentiment consistency analysis on the initial evaluation text to obtain the target analysis result corresponding to the initial evaluation text.

[0036] For example, the initial evaluation text is segmented to obtain multiple segmented texts. Then, a sentiment classification model such as the SVM model is used to classify the sentiment of each segmented text to obtain the sentiment type corresponding to each segmented text. All sentiment types are compared. When all sentiment types are completely consistent, the target analysis result corresponding to the initial evaluation text is determined to be sentiment consistent; when all sentiment types are not completely consistent, the target analysis result corresponding to the initial evaluation text is determined to be sentiment inconsistent.

[0037] In some implementations, the step of performing text sentiment consistency analysis on the initial evaluation text to obtain the target analysis result corresponding to the initial evaluation text includes: performing text segmentation on the initial evaluation text to obtain multiple segmented evaluation texts, and determining a first evaluation text and a first remaining evaluation text obtained by removing the first evaluation text from the multiple segmented evaluation texts; performing sentiment classification on the first evaluation text to obtain a first sentiment type corresponding to the first evaluation text and a first probability information corresponding to the first sentiment type; performing sentiment analysis on any second evaluation text in the first remaining evaluation text and the first evaluation text to obtain a second sentiment type corresponding to the second evaluation text and a second probability information corresponding to the second sentiment type; and comparing the first sentiment type and the second sentiment type to obtain the target analysis result. The type comparison results are obtained. When the type comparison results indicate that the first sentiment type and the second sentiment type are the same, a first analysis result corresponding to the first evaluation text and the second evaluation text is determined based on the first probability information and the second probability information. The first analysis result is used to characterize the analysis result of sentiment consistency between the first evaluation text and the second evaluation text. When the type comparison results indicate that the first sentiment type and the second sentiment type are not the same, the first analysis result corresponding to the first evaluation text and the second evaluation text is determined as a preset result. Based on the first analysis result, multiple segmented evaluation texts are merged to obtain a target merging result. Based on the target merging result, the initial evaluation text is subjected to text sentiment consistency analysis to obtain the target analysis result corresponding to the initial evaluation text.

[0038] For example, the initial evaluation text is segmented into multiple relatively independent sentences according to the language's grammatical rules, punctuation marks, and other segmentation methods to obtain multiple segmented evaluation texts.

[0039] For example, multiple segmented evaluation texts are sequentially identified as first evaluation texts, and then the selected first evaluation texts are removed from all segmented evaluation texts. The remaining segmented evaluation texts constitute the first residual evaluation text.

[0040] For example, a machine learning model, such as a Naive Bayes classifier or a support vector machine, is used to analyze the first evaluation text to obtain its corresponding first sentiment type (e.g., positive, negative, neutral) and the first probability information corresponding to that first sentiment type. For instance, if a machine learning model is used, the model will output the probability of each sentiment type, select the sentiment type with the highest probability as the first sentiment type, and record that probability as the first probability information.

[0041] For example, for each second evaluation text in the first remaining evaluation text, the second evaluation text and the first evaluation text are merged, i.e., text concatenation is performed. Sentiment analysis is then conducted using the same method as for sentiment classification of the first evaluation text, resulting in a second sentiment type and a second probability information corresponding to each second evaluation text. In other words, the second sentiment type combines the first and second evaluation texts. If the sentiment of the second evaluation text is consistent with that of the first evaluation text, the second probability information will be greater than the first probability information; that is, the second evaluation text plays an enhancing role. If the sentiment of the second evaluation text is inconsistent with that of the first evaluation text, or if the sentiment of the second evaluation text is relatively neutral, then the second probability information or the second sentiment type will be significantly different from the first probability information or the first sentiment type.

[0042] For example, the first sentiment type of the first evaluation text is compared one by one with the second sentiment type of each second evaluation text to obtain the type comparison result. When the type comparison result shows that the first sentiment type and the second sentiment type are the same, the first analysis result is determined by calculating the difference between their probabilities. For example, if the second probability information is greater than the first probability information, it indicates that the second evaluation text and the first evaluation text have a strong sentiment consistency, and the first analysis result can be directly determined as strongly consistent; if the second probability information is less than or equal to the first probability information, it indicates that the second evaluation text and the first evaluation text may have sentiment consistency or inconsistency, and the first analysis result can be directly determined as not necessarily consistent.

[0043] For example, when the type comparison result is that the first sentiment type and the second sentiment type are different, the first analysis result corresponding to the first evaluation text and the second evaluation text is determined as the preset result, and the preset result is set as sentiment inconsistency.

[0044] For example, the decision on which segmented evaluation texts can be merged is based on the results of the first analysis. For instance, if the first analysis result is strongly consistent, they can be merged directly; if the sentiment is inconsistent, they are not merged; if the first analysis result is not necessarily consistent, the second evaluation text needs to be classified as an independent text for sentiment classification. When the classification result is consistent with the sentiment classification result of the first evaluation text, the first and second evaluation texts are merged. Thus, multiple segmented evaluation texts are merged according to the above merging rules to obtain the target merged result.

[0045] For example, based on the target merging results, and considering the sentiment consistency of each merged text block, an overall sentiment consistency analysis is performed on the initial evaluation text. If there is only one merged result in the target merging results, it indicates that the target analysis result corresponding to the initial evaluation text is sentiment-consistent; if there are two or more merged results in the target merging results, it indicates that the target analysis result corresponding to the initial evaluation text is sentiment-inconsistent.

[0046] Specifically, when determining the first analysis result, not only are the sentiment types the same, but also the first and second probability information are considered. This makes the analysis results more objective and accurate. When the first and second sentiment types are different, the first analysis result is set as the preset result. This effectively identifies potentially contradictory sentiments in the text, thereby uncovering some hidden sentiment information in the text through text segmentation and sentiment consistency analysis. Then, based on the first analysis result, multiple segmented evaluation texts are merged to obtain the target merged result. Accurate text sentiment consistency analysis results can help businesses better understand consumers' true feelings, thus providing strong support for subsequent merchant value assessment.

[0047] In some implementations, the step of merging multiple segmented evaluation texts based on the first analysis result to obtain a target merging result includes: obtaining a first consistent text corresponding to the first evaluation text from the second evaluation text based on the first analysis result; removing the first consistent text from the first remaining evaluation text to obtain a second remaining evaluation text; performing sentiment classification on the first consistent text to obtain a third sentiment type corresponding to the first consistent text and third probability information corresponding to the third sentiment type; performing sentiment analysis on any third evaluation text in the second remaining evaluation text and the first consistent text to obtain a fourth sentiment type corresponding to the third evaluation text and fourth probability information corresponding to the fourth sentiment type; determining a second analysis result corresponding to the first consistent text and the third evaluation text based on the third sentiment type, the third probability information, the fourth sentiment type, and the fourth probability information; determining a fourth evaluation text from the second remaining evaluation text, and obtaining the first evaluation text and the fourth evaluation text from the first analysis result. The third analysis result corresponds to the first consistent text and the fourth evaluation text obtained from the second analysis result; when the third analysis result and the fourth analysis result are consistent, the first consistent text, the first evaluation text and the fourth evaluation text are merged according to the third analysis result to obtain a first merged result; when the third analysis result and the fourth analysis result are inconsistent, the first consistent text and the first evaluation text are merged to obtain a first merged text, and the sentiment consistency between the first merged text and the fourth evaluation text is calculated to obtain a fifth analysis result corresponding to the first merged text; the first consistent text, the first evaluation text and the fourth evaluation text are merged according to the fifth analysis result to obtain the first merged result; the first merged result is removed from the initial evaluation text to obtain other evaluation texts, and text merging is performed according to the other evaluation texts to obtain a second merged result; the target merged result corresponding to multiple segmented evaluation texts is determined according to the first merged result and the second merged result.

[0048] For example, a first analysis result is obtained between each second evaluation text and the first evaluation text. If the first analysis result indicates that the two are consistent in sentiment, these second evaluation texts are identified as first consistent texts corresponding to the first evaluation text. For instance, if the first analysis result shows that a certain second evaluation text has the same sentiment type as the first evaluation text and the second probability information is greater than the first probability information, then this second evaluation text belongs to the first consistent text category.

[0049] For example, the first consistent text is removed from the first remaining evaluation text, and the remaining text constitutes the second remaining evaluation text. This step is to separate the parts that are consistent with the sentiment of the first evaluation text so that they can be processed separately later.

[0050] For example, machine learning models, such as Naive Bayes classifiers and support vector machines, are used to analyze the first consistent text to obtain its corresponding sentiment type (e.g., positive, negative, neutral) and the probability information corresponding to that sentiment type, i.e., the third sentiment type and the third probability information. Then, for each third evaluation text in the second remaining evaluation text, the third evaluation text and the first consistent text are concatenated, and sentiment analysis is performed using the same method as for the sentiment classification of the first consistent text, resulting in a fourth sentiment type and the fourth probability information corresponding to each third evaluation text. That is, the fourth sentiment type combines the first consistent text and the third evaluation text. If the sentiment of the third evaluation text is consistent with that of the first consistent text, the fourth probability information will be greater than the third probability information, meaning the third evaluation text plays an enhancing role. If the sentiment of the third evaluation text is inconsistent with that of the first consistent text, or if the sentiment of the third evaluation text is relatively neutral, then the fourth probability information or the fourth sentiment type will be significantly different from the third probability information or the third sentiment type.

[0051] For example, a comparison is made between the third and fourth sentiment types to obtain a comparison result. When the comparison result shows that the third and fourth sentiment types are the same, if the third probability information is greater than the fourth probability information, then the second analysis result corresponding to the first consistent text and the third evaluation text is determined to be sentiment consistent. If the third probability information is less than or equal to the fourth probability information, then the second analysis result corresponding to the first consistent text and the third evaluation text is determined to be sentiment inconsistent. When the comparison result shows that the third and fourth sentiment types are different, then the second analysis result corresponding to the first consistent text and the third evaluation text is determined to be the preset result, that is, it is determined to be sentiment inconsistent.

[0052] For example, a fourth evaluation text is randomly selected multiple times from the second remaining evaluation texts. Then, the corresponding third analysis result between the first and fourth evaluation texts is found from the previously obtained first analysis result. Simultaneously, the corresponding fourth analysis result between the first consistent text and the fourth evaluation text is obtained from the second analysis result. When the third and fourth analysis results are consistent, it indicates that the first evaluation text, the first consistent text, and the fourth evaluation text have high sentiment consistency, and they can be merged into a first merged result. When the third and fourth analysis results are inconsistent, the first consistent text and the first evaluation text are first merged to obtain a first merged text. Then, the sentiment consistency between the first merged text and the fourth evaluation text is calculated to obtain a fifth analysis result. Finally, based on the fifth analysis result, it is decided whether to merge the fourth evaluation text with the first merged text to obtain the first merged result.

[0053] For example, the first merged result is removed from the initial evaluation text, and the remaining text is the other evaluation text. The sentiment analysis and merging steps described above are repeated for the other evaluation texts to obtain the second merged result.

[0054] For example, the first merging result and the second merging result are integrated to obtain the target merging result corresponding to multiple segmentation evaluation texts.

[0055] For example, if multiple segmented evaluation texts include text a, text b, text c, and text d, then text a is taken as the first evaluation text, and texts b, c, and d are determined as the first remaining evaluation texts. If text b is determined to be the first consistent text of the first evaluation text, then text b is determined as the first consistent text. Texts c and d are then determined as the second remaining evaluation texts. Text b is then subjected to sentiment classification to obtain a third sentiment type and the corresponding third probability information. Text b is then merged with texts c and d from the second remaining evaluation texts to obtain text b_c and text b_d, thereby obtaining the corresponding fourth sentiment type and the corresponding fourth probability information. The analysis results between texts b and c, and between texts b and d, are then combined to obtain the second analysis result corresponding to the first consistent text and the third evaluation text. Finally, any text from the second remaining evaluation texts is randomly selected as the fourth evaluation text. For example, text c is determined as the fourth evaluation text. The third analysis result between texts a and c is then obtained from the first analysis result, and the third analysis result from the second analysis result is obtained from the second analysis result. The results yield a fourth analysis result between text b and text c. If the third and fourth analysis results are consistent, and both indicate sentiment consistency, then text a, text b, and text c are directly merged to obtain the first merged result. If the third and fourth analysis results are consistent, but both indicate sentiment inconsistency, then text a and text b are merged, but not with text c, thus obtaining the first merged result. If the third and fourth analysis results are inconsistent, then the first consistent text (text b) and the first evaluation text (text a) are merged to obtain the first merged text (text a text b). The sentiment consistency between the first merged text and the fourth evaluation text (text c) is calculated using the aforementioned sentiment analysis method to obtain the fifth analysis result corresponding to the first merged text. If the fifth analysis result indicates sentiment consistency between the first merged text and the fourth evaluation text, then text a, text b, and text c are directly merged to obtain the first merged result. If the fifth analysis result indicates sentiment inconsistency between the first merged text and the fourth evaluation text, then text a and text b are merged, but not with text c, thus obtaining the first merged result.

[0056] Specifically, by performing sentiment classification and analysis on different subsets of text multiple times, it is possible to capture the emotional information in the text more meticulously. When dealing with the sentiment relationships between multiple texts, it is possible to effectively identify contradictory emotions that may exist in the text. Furthermore, through step-by-step text merging and analysis, it is possible to discover some hidden emotional information in the text. By merging texts with consistent sentiment, the accurate text merging results and sentiment analysis can help businesses better understand consumers' true feelings, thus providing strong support for subsequent merchant value assessment.

[0057] Step S103: Determine the target data status corresponding to the initial evaluation text based on the target analysis results and the initial scoring information.

[0058] For example, if the target analysis result is consistent with the sentiment of the initial evaluation text, then the target sentiment type corresponding to the initial evaluation text is obtained. Then, based on expert experience, the scoring range corresponding to the target sentiment type is obtained. Based on whether the initial scoring information is within the scoring range, the target data status corresponding to the initial evaluation text is determined. When the initial scoring information is within the scoring range, the target data status corresponding to the initial evaluation text is determined to be normal; when the initial scoring information is not within the scoring range, the target data status corresponding to the initial evaluation text is determined to be abnormal.

[0059] For example, if the target analysis result shows that the sentiment corresponding to the initial evaluation text is inconsistent, then the text proportion of each sentiment type in the initial evaluation text is obtained. Then, based on expert experience, the scoring range corresponding to each sentiment type is obtained. Then, the scoring range is weighted and summed according to the text proportion to obtain the target scoring range. Then, the target data status corresponding to the initial evaluation text is determined according to whether the initial scoring information is within the target scoring range. When the initial scoring information is within the target scoring range, the target data status corresponding to the initial evaluation text is determined to be normal; when the initial scoring information is not within the target scoring range, the target data status corresponding to the initial evaluation text is determined to be abnormal.

[0060] In some implementations, determining the target data state corresponding to the initial evaluation text based on the target analysis result and the initial rating information includes: when the target analysis result is the preset result, determining the first position distribution information corresponding to the first merged result and the second position distribution information corresponding to the second merged result based on the initial evaluation text; determining the target user's core evaluation text and target intent corresponding to the initial evaluation text based on the first position distribution information and the second position distribution information; determining the rating interval corresponding to the initial evaluation text based on the target intent and the mapping table; and determining the target data state corresponding to the initial evaluation text based on the rating interval and the initial rating information.

[0061] For example, it is determined whether the target analysis result is a preset result. That is, it is determined whether the target analysis result is inconsistent in sentiment. When the target analysis result is inconsistent in sentiment, the first position distribution information of the first merged result in the initial evaluation text and the second position distribution information of the second merged result in the initial evaluation text are determined based on the initial evaluation text. The first position distribution information or the second position distribution information can be represented by sentence number or character position.

[0062] For example, by combining the first position distribution information and the second position distribution information, it is determined which of the first and second merged results in the initial evaluation text is more often in positions such as the beginning and the end. Then, based on the first and second position distribution information of the first and second merged results, the sentiment tendency and semantic expression of the initial evaluation text are determined, thereby obtaining the target intent corresponding to the target user.

[0063] For example, if the text of the first merged result focuses on the beginning and expresses high praise for the product, while the text of the second merged result raises some minor questions at the end, then the text in the beginning is likely the core evaluation text, and the intended purpose may be to emphasize the product's advantages.

[0064] For example, a mapping table is created that records the correspondence between different target intentions and rating ranges. For instance, a target intention of "high praise" may correspond to a rating range of 8-10 points; "moderately satisfied with minor issues" corresponds to 6-8 points; and "dissatisfied" corresponds to 1-3 points, etc. Thus, based on the determined target intention, the corresponding rating range can be found in the mapping table.

[0065] For example, the determined rating range is compared with the initial rating information. The initial rating information may be a specific score given by the user. If the initial rating is within the rating range, it indicates that the evaluation is relatively reasonable, and the target data status can be marked as "normal"; if the initial rating is higher than the upper limit of the rating range or lower than the lower limit of the rating range, the target data status can be marked as "abnormal".

[0066] Specifically, by analyzing the location distribution information of the text to determine the core evaluation text and target intent, a more comprehensive and accurate understanding of users' true evaluations can be achieved. This avoids judging the reasonableness of an evaluation solely based on a single initial rating, takes into account the richness and complexity of the text content, and allows for a more in-depth analysis of the evaluations. Furthermore, by using a mapping table to map target intent to rating ranges, an objective standard for evaluations is provided. This reduces the impact of inconsistent rating scales among different users, improving the comparability and reliability of evaluations.

[0067] Step S104: Calculate the information correlation degree between the historical consumption information and the initial evaluation text.

[0068] For example, based on the target user's historical consumption information, such as order records from online shopping platforms or offline purchases, the type of product the target user consumes can be determined. The product type can be a specific product from the target merchant.

[0069] For example, natural language processing techniques, such as word segmentation and part-of-speech tagging, are used to extract words related to consumer reviews from the initial review text, such as "good quality," "affordable price," and "attentive service." These words are the consumer review terms. Then, for each type of consumer product, the corresponding consumer review terms are associated. For example, if the consumer product type is an electronic product, the consumer review terms might be "powerful performance" and "clear screen." Statistical analysis can then be used to identify the associations between different consumer product types and consumer review terms, such as which review terms appear most frequently in the review text of a specific product type. The extracted consumer review terms are then matched with the identified consumer product types to determine which product type each review term is associated with. Furthermore, the frequency of the review terms in the review text of a specific product type determines the corresponding association index, thus calculating a score for the association between each pair of consumer product types and consumer review terms, thereby obtaining the information correlation degree between historical consumer information and the initial review text.

[0070] For example, the above method can reduce malicious reviews of target merchants by peers, thereby improving the accuracy and reliability of subsequent comprehensive merchant value assessments of target merchants.

[0071] In some implementations, calculating the information correlation between the historical consumption information and the initial review text includes: performing part-of-speech analysis on the initial review text to obtain relevant nouns and evaluation words corresponding to relevant products in the initial review text; determining preset evaluation words corresponding to the relevant nouns, and calculating the correlation between the relevant nouns and the relevant evaluation words based on the preset evaluation words; determining the association relationship between the relevant nouns and the relevant evaluation words based on the correlation; filtering the relevant evaluation words based on the association relationship to obtain remaining evaluation words, and determining related products that can be used to evaluate the target merchant based on the remaining evaluation words; obtaining the target consumption product corresponding to the target user from the historical consumption information; calculating the overlap degree based on the target consumption product, the relevant nouns, and the related products to obtain a target overlap degree; and determining the information correlation between the historical consumption information and the initial review text based on the target overlap degree.

[0072] For example, the initial evaluation text can be processed using tools such as LTP from Harbin Institute of Technology and jieba word segmentation to label each word in the initial evaluation text with its corresponding part of speech, such as noun, verb, adjective, etc. Then, product-related nouns, such as "mobile phone", "clothes", "cosmetics", etc., can be filtered out from the initial evaluation text with labeled parts of speech. At the same time, relevant evaluation words used to describe product features, quality, and user experience, such as "easy to use", "beautiful", "durable", etc., can be extracted.

[0073] For example, a pre-built evaluation term library can be established for different types of products. For instance, for mobile phones, pre-built evaluation terms might include "good performance," "clear camera," and "long battery life." Then, statistical methods can be used to calculate the correlation between related nouns and evaluation terms. For example, the frequency of a related evaluation term in texts associated with that noun can be calculated; the higher the frequency, the higher the correlation. Alternatively, semantic similarity methods can be used to determine the degree of semantic similarity between related evaluation terms and pre-built evaluation terms, thereby determining the correlation.

[0074] For example, a correlation threshold is set. When the correlation between related nouns and related evaluation words exceeds this threshold, they are considered to be related. Based on the set threshold, it is determined which related evaluation words are related to related nouns, and these related combinations are recorded.

[0075] For example, based on the established associations, relevant evaluation terms are filtered to remove those that are not related to the relevant nouns, resulting in the remaining evaluation terms. Then, based on the characteristics and attributes described by the remaining evaluation terms, combined with the relevant nouns, related products that can be used to evaluate the target merchant are determined. For instance, if the remaining evaluation terms are "soft" and "breathable," and the relevant noun is "towel," then the related product can be determined to be the towel product of the target merchant.

[0076] For example, the historical consumption information of the target user is organized and analyzed to extract the product information consumed by the target user and determine the target consumption products. Then, the target consumption products, related terms, and associated products are compared to calculate their overlap. For example, the target overlap can be obtained by calculating the proportion of identical products to the total number of products. Based on the calculated target overlap, the information correlation between historical consumption information and the initial evaluation text is determined. The higher the target overlap, the higher the information correlation between the two.

[0077] Specifically, to accurately identify malicious or erroneous reviews of target merchants in the initial evaluation text, the authenticity and rationality of the reviews can be discovered by correlating the target user's historical consumption information with the initial evaluation text. This correlation analysis can effectively eliminate such unreasonable reviews, reduce the impact of malicious reviews on target merchants, and, after reducing malicious or erroneous reviews, the data in subsequent user-merchant comprehensive value assessments can more accurately reflect the actual operating conditions and service quality of target merchants, thus making the assessment results more credible and significantly improving the accuracy and reliability of merchant comprehensive value assessments.

[0078] Step S105: Delete the initial evaluation text according to the target data status and the information correlation to obtain the target evaluation text, and obtain the associated rating information corresponding to the target evaluation text from the initial rating information.

[0079] For example, if the target data status corresponding to the initial evaluation text is abnormal or the information correlation between the initial evaluation text and historical consumption information is low, the initial evaluation text is deleted to obtain the target evaluation text.

[0080] For example, the associated rating information is obtained by obtaining the rating information corresponding to each target evaluation text from the initial rating information.

[0081] Step S106: Perform data clustering on the target evaluation text to obtain target clustering results, and perform data correction on the associated rating information based on the target clustering results to obtain target rating information corresponding to the target evaluation text.

[0082] For example, multiple target keywords are extracted from the target evaluation text, and then the word2vec model is used to obtain the text vectors corresponding to the target keywords. The text vectors are then used to perform text clustering on the target evaluation text using the k-means clustering algorithm to obtain the target clustering results.

[0083] For example, relevant rating information corresponding to each sub-cluster in the target clustering result is obtained from the associated rating information, and anomaly identification is performed on the relevant rating information to obtain abnormal rating data. Then, the abnormal rating data is corrected according to the normal rating data in the sub-cluster so that the abnormal rating data is consistent with the normal rating data, thereby obtaining the target rating information corresponding to each target evaluation text.

[0084] For example, the mean of relevant rating information within each sub-cluster is calculated. If the relevant rating information of a text differs significantly from the mean of its sub-cluster, its rating can be adjusted to the mean of that cluster. For instance, if the mean rating within a cluster is 8 points, and one text has a rating of only 3 points, but this text is semantically similar to other texts in the cluster, then its rating can be adjusted to 8 points. Thus, based on the determined adjustment strategy, the associated rating information is corrected. The rating of each target evaluation text is adjusted to the corrected value, thereby obtaining the target rating information corresponding to the target evaluation text.

[0085] Specifically, by performing cluster analysis on the target evaluation text, outliers and biases in the scoring can be identified and corrected. This allows the scoring to more accurately reflect the actual situation expressed in the text, providing a more reliable basis for subsequent decision-making.

[0086] In some implementations, the step of clustering the target evaluation text to obtain target clustering results includes: selecting any text from the target evaluation text as the current evaluation text, and calculating the text similarity between any text in the target evaluation text and the current evaluation text; determining the target probability corresponding to the target evaluation text being selected as a cluster center based on the text similarity; determining the initial cluster center corresponding to the target evaluation text at a preset number based on the target probability, and clustering the target evaluation text based on the initial cluster center to obtain initial clustering results; determining the first text representation vector corresponding to the initial cluster center based on a text representation model, and determining the second text representation vector corresponding to each sub-text of each first sub-cluster in the initial clustering results based on the text representation model; performing vector decomposition on the first text representation vector to obtain multiple first sub-vectors and performing vector decomposition on the second text representation vector. The process involves: decomposing the first and second sub-vectors to obtain multiple second sub-vectors; processing the first and second sub-vectors for mean and standard deviation to obtain a first mean and a first standard deviation; performing a count on each first sub-cluster in the initial clustering result to obtain the number of texts corresponding to the first sub-cluster; determining the cluster weights corresponding to the first sub-clusters based on the number of texts; and determining the first membership probability of any text in the target evaluation text under the first sub-cluster based on the first mean and the first standard deviation combined with a Gaussian model; adjusting the first mean and the first standard deviation based on the cluster weights combined with the first membership probability to obtain a second mean and a second standard deviation; determining the second membership probability of any text in the target evaluation text under the first sub-cluster based on the second mean and the second standard deviation combined with the Gaussian model; and performing data clustering on the target evaluation text based on the second membership probability to obtain the target clustering result.

[0087] For example, a text is randomly selected from the target evaluation text as the current evaluation text, and then the similarity between each of the remaining texts in the target evaluation text and the current evaluation text is calculated according to the text similarity calculation method such as cosine similarity, edit distance, etc., so as to obtain the text similarity.

[0088] For example, based on the calculated text similarity, the target probability of the target evaluation text being selected as a cluster center is determined. The lower the similarity of the text to the current evaluation text, the higher the probability of it being selected as a cluster center. Therefore, based on a preset number of cluster centers, a corresponding number of texts are selected as initial cluster centers according to the target probability from low to high.

[0089] For example, based on the selected initial cluster center, each text in the target evaluation text is assigned to the cluster containing the initial cluster center that is most similar to it, thereby obtaining the initial clustering result, and each cluster is a first sub-cluster.

[0090] For example, using text representation models such as word vector models (Word2Vec, GloVe) or pre-trained language models (BERT, RoBERTa), the initial cluster centers are converted into first text representation vectors, while each subtext in each first sub-cluster of the initial clustering results is converted into a second text representation vector.

[0091] For example, a vector decomposition operation is performed on the first and second text representation vectors, such as using principal component analysis. This decomposes the first text representation vector into first sub-vectors corresponding to multiple dimensions, and the second text representation vector into multiple second sub-vectors corresponding to multiple dimensions. The mean and standard deviation of the first and second sub-vectors under different dimensions are then calculated to obtain the first mean and first standard deviation for each dimension. The first mean reflects the average level of the first and second sub-vectors under the relevant dimensions, while the first standard deviation reflects the degree of dispersion between the first and second sub-vectors under the relevant dimensions.

[0092] For example, the number of texts contained in each first sub-cluster in the initial clustering results is counted. The cluster weight of each first sub-cluster is then determined based on the counted number of texts; generally, the more texts a sub-cluster contains, the greater its cluster weight. The cluster weight can be obtained by dividing the number of texts in each sub-cluster by the total number of texts in all sub-clusters.

[0093] For example, a corresponding target model is constructed based on the first mean and the first standard deviation combined with a Gaussian model. The second sub-vector corresponding to the target evaluation text is then substituted into the target model to obtain the first membership probability of any text in the target evaluation text under each first sub-cluster.

[0094] For example, after obtaining the first membership probability of any text in the target evaluation text under each first sub-cluster, and then combining it with the cluster weights for weighted summation and normalization, the target membership probability of any text in the target evaluation text under each first sub-cluster is obtained. This target membership probability is then combined with a second sub-vector to adjust the first standard deviation or first mean to obtain the second standard deviation or second mean. For example, the second standard deviation or second mean is obtained according to the following formula:

[0095] ;

[0096] ;

[0097] in, This represents the second mean of the i-th sub-cluster under the w-th dimension. This represents the number of texts corresponding to the i-th sub-cluster. This represents the target membership probability corresponding to the k-th text in the i-th sub-cluster under the w-th dimension. This represents the second sub-vector corresponding to the k-th text of the i-th sub-cluster under the w-th dimension. This represents the second standard deviation corresponding to the i-th sub-cluster under the w-th dimension. This represents the average of the second mean values ​​corresponding to the i-th sub-cluster under the w-th dimension.

[0098] For example, the target model is reconstructed based on the second mean and the second standard deviation combined with the Gaussian model. Then, the text vector corresponding to any text in the target evaluation text under the target dimension is substituted into the target model corresponding to each sub-cluster. This yields the second membership probability of any text in the target evaluation text under each first sub-cluster in the target dimension. Based on the calculated second membership probability, the final membership probability between each text in the target evaluation text and the first sub-cluster in all target dimensions is calculated. Finally, each text in the target evaluation text is reassigned to the first sub-cluster with the highest final membership probability, completing the final data clustering and obtaining the target clustering result.

[0099] Specifically, in the clustering process, by processing text vectors and calculating membership probabilities, similar texts can be grouped together more accurately, thereby improving the quality and reliability of the data.

[0100] In some implementations, calculating the text similarity between any text in the target evaluation text and the current evaluation text includes: determining any text in the target evaluation text as the text to be processed, and performing keyword recognition on the text to be processed to obtain a first keyword; performing keyword recognition on the current evaluation text to obtain a second keyword; converting the first keyword to a target space to obtain a first feature representation corresponding to the target dimension, and converting the second keyword to the target space to obtain a second feature representation corresponding to the target dimension; determining the first distribution information corresponding to the first keyword in the target space based on the first feature representation, and determining the first entropy value corresponding to the first keyword in the target space based on the first feature representation; determining the second distribution information corresponding to the second keyword in the target space based on the second feature representation, and determining the second entropy value corresponding to the second keyword in the target space based on the second feature representation; determining the first separation degree corresponding to the first keyword in the target dimension based on the first distribution information and the first entropy value, wherein the first separation degree is used to characterize the first... The expression intensity of the keyword in the text to be processed; the second separation degree corresponding to the second keyword in the target dimension is determined according to the second distribution information and the second entropy value, the second separation degree is used to characterize the expression intensity of the second keyword in the current evaluation text; the first dimension weight corresponding to the first keyword in the target dimension is determined, and the first frequency of the first keyword appearing in the first similar text cluster corresponding to the current evaluation text is determined; the first similarity between the first keyword and the current evaluation text is determined according to the first dimension weight, the first frequency and the first separation degree; the second dimension weight corresponding to the second keyword in the target dimension is determined, and the second frequency of the second keyword appearing in the second similar text cluster corresponding to the text to be processed is determined; the second similarity between the second keyword and the text to be processed is determined according to the second dimension weight, the second frequency and the second separation degree; the text similarity between the text to be processed and the current evaluation text is determined by fusing the first similarity and the second similarity; wherein, the first similarity is obtained according to the following formula:

[0101] ;

[0102] in, This represents the k-th keyword of the i-th text to be processed. With the j-th current evaluation text The first similarity between them, where count represents the number corresponding to the target dimension. The frequency of the first keyword appearing in the first similar text cluster corresponding to the i-th text to be processed is represented by the first frequency. and Represents a constant. This represents the i-th text to be processed. The corresponding text length, This represents the average text length of the first similar text cluster corresponding to the j-th currently evaluated text.

[0103] For example, one text is selected from the target evaluation text as the text to be processed, and a keyword extraction algorithm (such as TF-IDF, TextRank, etc.) is used to process it to identify important words and obtain the first keyword. The same keyword extraction algorithm is used to process the current evaluation text to obtain the second keyword.

[0104] For example, a target space is selected, which can be a pre-trained word vector space. The first keyword is then mapped into this target space to obtain the first feature representation corresponding to the target dimension. This first feature representation describes the position of the first keyword in the target space in vector form. The second keyword is transformed into the target space in the same way to obtain the second feature representation corresponding to the target dimension.

[0105] For example, based on the first feature representation, the distribution of the first keyword in the target space is analyzed to obtain first distribution information, such as the probability distribution of the keyword in various dimensions. Then, based on the first feature representation, the first entropy value of the first keyword in the target space is calculated. The entropy value can measure the uncertainty of the keyword distribution; the larger the entropy value, the more dispersed the keyword distribution. Following the same method, the second distribution information and second entropy value of the second keyword in the target space are determined based on the second feature representation.

[0106] For example, combining the first distribution information and the first entropy value, the first separation degree of the first keyword in the target dimension is calculated. The first separation degree reflects the uniqueness and expressive intensity of the first keyword in the text to be processed; the higher the separation degree, the more important the keyword is in the text to be processed. Then, based on the second distribution information and the second entropy value, the second separation degree of the second keyword in the target dimension is calculated to characterize the expressive intensity of the second keyword in the current evaluation text.

[0107] For example, a first-dimensional weight is determined for the first keyword under the target dimension. This weight can be determined based on factors such as the keyword's importance in the target space and its contribution to text classification. Then, the first frequency of the first keyword appearing in the first similar text cluster corresponding to the current evaluation text is calculated. The first similar text cluster can be a set of texts similar to the current evaluation text obtained from a database using a clustering algorithm. Similarly, a second-dimensional weight for the second keyword under the target dimension is determined, and its second frequency appearing in the second similar text cluster corresponding to the text to be processed is calculated.

[0108] For example, the first similarity between the first keyword and the current evaluation text is calculated by combining the first dimension weight, the first frequency, and the first separation degree according to the following formula:

[0109] ;

[0110] in, This represents the k-th first keyword of the i-th text to be processed. With the j-th current evaluation text The first similarity between them, where count represents the number of items corresponding to the target dimension. This represents the frequency at which the k-th first keyword of the i-th text to be processed appears in the first similar text cluster corresponding to the h-th current evaluation text. and Represents a constant. This represents the i-th text to be processed. The corresponding text length, This represents the average text length of the first closely related text cluster corresponding to the j-th current evaluation text.

[0111] For example, the second similarity between the second keyword and the text to be processed is calculated in the same way based on the second dimension weight, the second frequency, and the second separation degree.

[0112] For example, after obtaining the first similarity between the first keyword and the current evaluation text according to the above method, the first similarity between each first keyword and the current evaluation text is obtained according to the above method. Then, all the first similarities are summed and averaged to obtain the first total similarity between the text to be processed and the current evaluation text. Then, the second similarity between the second keyword and the text to be processed is obtained according to the above method. Then, each second keyword and the text to be processed is obtained according to the above method. Then, all the second similarities are summed and averaged to obtain the second total similarity between the text to be processed and the current evaluation text. Finally, the first total similarity and the second total similarity are averaged to obtain the text similarity between the text to be processed and the current evaluation text.

[0113] Specifically, the above method comprehensively considers multiple factors such as keyword distribution information, entropy value, separation degree, dimensional weight, and frequency. This allows the calculated text similarity to more accurately reflect the semantic relevance between texts, reducing misjudgments caused by superficially similar but semantically different words, thereby improving the reliability and accuracy of text similarity.

[0114] In some implementations, the step of correcting the associated rating information based on the target clustering results to obtain the target rating information corresponding to the target evaluation text includes: obtaining the relevant rating information corresponding to each second sub-cluster in the target clustering results from the associated rating information; identifying abnormal data in the relevant rating information to obtain abnormal rating information and the relevant evaluation text corresponding to the abnormal rating information; removing the abnormal rating information from the relevant rating information to obtain the remaining rating information; and correcting the rating of the relevant evaluation text based on the remaining rating information to obtain the target rating information.

[0115] For example, based on the previously obtained target clustering results, each second sub-cluster in the target clustering results is associated with the associated scoring information. The relevant scoring information corresponding to each second sub-cluster is then extracted from the associated scoring information. This means extracting the scoring data related to each specific cluster separately for subsequent processing of each sub-cluster.

[0116] For example, the extracted relevant rating information can be analyzed using anomaly data identification methods, such as machine learning-based methods like the Isolation Forest algorithm, which can identify isolated outliers in the dataset that differ from most data patterns, thereby obtaining the relevant evaluation text corresponding to the anomaly rating information.

[0117] For example, the identified abnormal rating information is removed from the relevant rating information to obtain the remaining rating information. This step is to ensure that the data used for subsequent rating correction is relatively reliable and conforms to the overall pattern, avoiding the adverse effects of abnormal data on the final rating result. Therefore, based on the mean, median, and other statistical measures of the remaining rating information, it is used as the corresponding rating result after the relevant evaluation text in this sub-category is modified, thereby obtaining the corrected target rating information.

[0118] Specifically, by removing abnormal rating information, the reliability and stability of the rating data are improved, so that the corrected target rating information is more in line with the actual situation.

[0119] Step S107: Based on the historical consumption information and the target evaluation text, combined with the target rating information, use an artificial intelligence model to determine the merchant value assessment result corresponding to the target merchant.

[0120] For example, historical merchant consumption information, review texts, and rating information are obtained from a database. These consumer information features, review text vectors, and rating information are then fused to combine features from different sources into a single feature vector. Alternatively, feature fusion can be weighted according to their importance, highlighting the role of certain key features. Then, artificial intelligence models, such as machine learning models (e.g., decision trees, random forests, support vector machines) or deep learning models (e.g., multilayer perceptrons, long short-term memory networks (LSTM), etc.), are used as input to train the selected model, with known merchant value assessment results as output. During training, the model parameters are adjusted to minimize the error between the predicted and actual results, thereby optimizing the model's performance.

[0121] For example, a fused feature vector is obtained by combining historical consumption information, target review text, and target rating information corresponding to the target merchant. This fused feature vector is then input into a trained artificial intelligence model. The model calculates based on learned patterns and rules, outputting a merchant value assessment result for the target merchant. This result can be a specific numerical value or a classification label (such as high value, medium value, low value, etc.). Thus, the target merchant can clearly understand its market evaluation through the merchant value assessment result and optimize and improve itself accordingly.

[0122] Please see Figure 2 , Figure 2This application provides a rating modification system 200 for review texts. The system includes a data acquisition module 201, a sentiment analysis module 202, a state determination module 203, a correlation calculation module 204, a data correlation module 205, a data correction module 206, and a value assessment module 207. The data acquisition module 201 obtains historical consumption information of a target user after making a purchase at a target merchant, and obtains the initial review text and initial rating information corresponding to the initial review text under the historical consumption information. The sentiment analysis module 202 performs text sentiment consistency analysis on the initial review text to obtain the target analysis result corresponding to the initial review text. The state determination module 203 determines the value of the review text based on the target analysis result and the initial rating information. The system comprises: an initial evaluation text corresponding to the target data status; an association calculation module 204, used to calculate the information correlation degree between the historical consumption information and the initial evaluation text; a data association module 205, used to delete the initial evaluation text according to the target data status and the information correlation degree to obtain the target evaluation text, and to obtain the associated rating information corresponding to the target evaluation text from the initial rating information; a data correction module 206, used to perform data clustering on the target evaluation text to obtain the target clustering result, and to perform data correction on the associated rating information according to the target clustering result to obtain the target rating information corresponding to the target evaluation text; and a value assessment module 207, used to determine the merchant value assessment result corresponding to the target merchant using an artificial intelligence model based on the historical consumption information, the target evaluation text, and the target rating information.

[0123] In some implementations, the rating modification system 200 for the evaluation text can be applied to a terminal device.

[0124] It should be noted that those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the evaluation text scoring modification system 200 described above can be referred to the corresponding process in the aforementioned evaluation text scoring modification method embodiment, and will not be repeated here.

[0125] Please see Figure 3 , Figure 3 This is a schematic block diagram of the structure of a terminal device provided in an embodiment of the present invention.

[0126] like Figure 3 As shown, the terminal device 300 includes a processor 301 and a memory 302, which are connected via a bus 303, such as an I2C (Inter-integrated Circuit) bus.

[0127] Specifically, processor 301 provides computing and control capabilities to support the operation of the entire terminal device. Processor 301 can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0128] Specifically, the memory 302 can be a Flash chip, a read-only memory (ROM) disk, an optical disk, a USB flash drive, or a portable hard drive, etc.

[0129] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the embodiments of the present invention, and does not constitute a limitation on the terminal device to which the embodiments of the present invention are applied. A specific server may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0130] The processor is used to run a computer program stored in the memory, and when executing the computer program, implements any of the rating modification methods for evaluation text provided in the embodiments of the present invention.

[0131] In one embodiment, the processor is configured to run a computer program stored in memory, and when executing the computer program, perform the following steps:

[0132] Obtain historical consumption information of the target user after making a purchase at the target merchant, and obtain the initial evaluation text of the target merchant under the historical consumption information and the initial rating information corresponding to the initial evaluation text;

[0133] Perform text sentiment consistency analysis on the initial evaluation text to obtain the target analysis results corresponding to the initial evaluation text;

[0134] Based on the target analysis results and the initial scoring information, determine the target data status corresponding to the initial evaluation text;

[0135] Calculate the information correlation degree between the historical consumption information and the initial evaluation text;

[0136] Based on the target data status and the information correlation, the initial evaluation text is deleted to obtain the target evaluation text, and the associated rating information corresponding to the target evaluation text is obtained from the initial rating information;

[0137] The target evaluation text is clustered to obtain target clustering results, and the associated rating information is corrected according to the target clustering results to obtain target rating information corresponding to the target evaluation text.

[0138] Based on the historical consumption information, the target evaluation text, and the target rating information, an artificial intelligence model is used to determine the merchant value assessment result corresponding to the target merchant.

[0139] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the terminal device described above can be referred to the corresponding process in the aforementioned embodiment of the method for modifying the rating of evaluation text, and will not be repeated here.

[0140] This invention also provides a storage medium for computer-readable storage, the storage medium storing one or more programs, which can be executed by one or more processors to implement the steps of any of the rating modification methods for evaluation text provided in the specification of this invention.

[0141] The storage medium can be an internal storage unit of the terminal device described in the aforementioned embodiments, such as a hard drive or memory of the terminal device. Alternatively, the storage medium can be an external storage device of the terminal device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card.

[0142] It will be understood by those skilled in the art that all or some of the steps, systems, or apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware embodiments, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as microprocessors, such as central processing units, digital signal processors, or software executed by microprocessors, or as hardware, or as integrated circuits, such as application-specific integrated circuits (ASICs). Such software may be distributed on computer-readable media, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0143] It should be understood that the term "and / or" as used in this specification and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. It should be noted that, herein, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0144] The sequence numbers of the above embodiments of the present invention are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. The above are only specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for modifying the rating of evaluation text, characterized in that, The method includes: Obtain historical consumption information of the target user after making a purchase at the target merchant, and obtain the initial evaluation text of the target merchant under the historical consumption information and the initial rating information corresponding to the initial evaluation text; The initial evaluation text is segmented to obtain multiple segmented evaluation texts, and a first evaluation text is determined from the multiple segmented evaluation texts, and a first remaining evaluation text is obtained by removing the first evaluation text from the multiple segmented evaluation texts. Perform sentiment classification on the first evaluation text to obtain the first sentiment type corresponding to the first evaluation text and the first probability information corresponding to the first sentiment type; After merging any second evaluation text in the first remaining evaluation text with the first evaluation text, sentiment analysis is performed to obtain the second sentiment type corresponding to the second evaluation text and the second probability information corresponding to the second sentiment type. The first sentiment type and the second sentiment type are compared to obtain a type comparison result. When the type comparison result is that the first sentiment type and the second sentiment type are the same, the first analysis result corresponding to the first evaluation text and the second evaluation text is determined according to the first probability information and the second probability information. The first analysis result is used to characterize the analysis result of the sentiment consistency between the first evaluation text and the second evaluation text. When the type comparison result is that the first emotion type and the second emotion type are not the same, the first analysis result corresponding to the first evaluation text and the second evaluation text is determined as the preset result; Based on the first analysis result, multiple segmentation evaluation texts are merged to obtain the target merging result; Based on the target merging results, perform text sentiment consistency analysis on the initial evaluation text to obtain the target analysis results corresponding to the initial evaluation text; Based on the target analysis results and the initial scoring information, determine the target data status corresponding to the initial evaluation text; Calculate the information correlation degree between the historical consumption information and the initial evaluation text; Based on the target data status and the information correlation, the initial evaluation text is deleted to obtain the target evaluation text, and the associated rating information corresponding to the target evaluation text is obtained from the initial rating information; The target evaluation text is clustered to obtain target clustering results, and the associated rating information is corrected based on the target clustering results to obtain the target rating information corresponding to the target evaluation text.

2. The method according to claim 1, characterized in that, The step of merging multiple segmented evaluation texts based on the first analysis result to obtain the target merging result includes: Based on the first analysis result, obtain the first consistent text corresponding to the first evaluation text from the second evaluation text; The first consistent text is removed from the first remaining evaluation text to obtain the second remaining evaluation text; Sentiment classification is performed on the first consistent text to obtain the corresponding third sentiment type and the third probability information corresponding to the third sentiment type; After merging any third evaluation text in the second remaining evaluation text with the first consistent text, sentiment analysis is performed to obtain the fourth sentiment type corresponding to the third evaluation text and the fourth probability information corresponding to the fourth sentiment type. The second analysis result corresponding to the first consistent text and the third evaluation text is determined based on the third sentiment type, the third probability information, the fourth sentiment type, and the fourth probability information; Determine the fourth evaluation text from the second remaining evaluation text, and obtain the third analysis result corresponding to the first evaluation text and the fourth evaluation text from the first analysis result, and obtain the fourth analysis result corresponding to the first consistent text and the fourth evaluation text from the second analysis result; When the third analysis result and the fourth analysis result are consistent, the first consistent text, the first evaluation text and the fourth evaluation text are merged according to the third analysis result to obtain the first merged result; When the third analysis result and the fourth analysis result are inconsistent, the first consistent text and the first evaluation text are merged to obtain the first merged text, and the sentiment consistency between the first merged text and the fourth evaluation text is calculated to obtain the fifth analysis result corresponding to the first merged text. Based on the fifth analysis result, the first consistency text, the first evaluation text, and the fourth evaluation text are merged to obtain the first merging result; The first merging result is removed from the initial evaluation text to obtain other evaluation texts, and text merging is performed based on the other evaluation texts to obtain a second merging result; The target merging result corresponding to multiple segmentation evaluation texts is determined based on the first merging result and the second merging result.

3. The method according to claim 2, characterized in that, The step of determining the target data status corresponding to the initial evaluation text based on the target analysis results and the initial scoring information includes: When the target analysis result is the preset result, the first position distribution information corresponding to the first merged result is determined according to the initial evaluation text, and the second position distribution information corresponding to the second merged result is determined according to the initial evaluation text. Based on the first location distribution information and the second location distribution information, the target user's core evaluation text corresponding to the initial evaluation text and the target intent corresponding to the core evaluation text are determined. The scoring range corresponding to the initial evaluation text is determined based on the target intent and the mapping table, and the target data state corresponding to the initial evaluation text is determined based on the scoring range and the initial scoring information.

4. The method according to claim 1, characterized in that, The calculation of the information correlation between the historical consumption information and the initial evaluation text includes: Part-of-speech analysis is performed on the initial evaluation text to obtain relevant nouns and evaluation terms corresponding to the relevant products in the initial evaluation text; Determine the preset evaluation words corresponding to the relevant nouns, and calculate the correlation between the relevant nouns and the relevant evaluation words based on the preset evaluation words; The correlation between the relevant nouns and the relevant evaluation terms is determined based on the correlation degree. Based on the aforementioned association, the relevant evaluation terms are filtered to obtain the remaining evaluation terms, and based on the remaining evaluation terms, the associated products that can be used to evaluate the target merchant are determined. The target consumption product corresponding to the target user is obtained from the historical consumption information; The target overlap is obtained by calculating the overlap based on the target consumer product, the relevant terms, and the associated products. The degree of information correlation between the historical consumption information and the initial evaluation text is determined based on the target overlap.

5. The method according to claim 1, characterized in that, The step of performing data clustering on the target evaluation text to obtain the target clustering result includes: Select any text from the target evaluation texts as the current evaluation text, and calculate the text similarity between any text in the target evaluation texts and the current evaluation text; The target probability corresponding to the selection of the target evaluation text as a cluster center is determined based on the text similarity. Based on the target probability, the initial cluster centers corresponding to the target evaluation text under a preset number are determined, and the target evaluation text is clustered based on the initial cluster centers to obtain the initial clustering results; The first text representation vector corresponding to the initial cluster center is determined according to the text representation model, and the second text representation vector corresponding to each subtext of each first sub-cluster in the initial clustering result is determined according to the text representation model. The first text representation vector is decomposed into multiple first sub-vectors, and the second text representation vector is decomposed into multiple second sub-vectors. The first mean and the first standard deviation are obtained by processing the first sub-vector and the second sub-vector according to the mean and standard deviation. The number of texts corresponding to each of the first sub-clusters in the initial clustering results is obtained by counting the number of texts in each first sub-cluster. The cluster weight corresponding to the first sub-cluster is determined based on the number of texts, and the first membership probability of any text in the target evaluation text under the first sub-cluster is determined based on the first mean and the first standard deviation combined with a Gaussian model. The first mean and the first standard deviation are adjusted according to the cluster weights and the first membership probability to obtain the second mean and the second standard deviation; Based on the second mean and the second standard deviation, combined with the Gaussian model, determine the second membership probability of any text in the target evaluation text under the first sub-cluster; The target evaluation text is clustered based on the second membership probability to obtain the target clustering result.

6. The method according to claim 5, characterized in that, The calculation of the text similarity between any text in the target evaluation text and the current evaluation text includes: Identify any one of the target evaluation texts as the text to be processed, and perform keyword recognition on the text to be processed to obtain the first keyword; The second keyword is obtained by keyword recognition of the current evaluation text; The first keyword is converted into the target space to obtain the first feature representation corresponding to the target dimension, and the second keyword is converted into the target space to obtain the second feature representation corresponding to the target dimension. Based on the first feature representation, determine the first distribution information corresponding to the first keyword in the target space, and based on the first feature representation, determine the first entropy value corresponding to the first keyword in the target space; The second distribution information corresponding to the second keyword in the target space is determined based on the second feature representation, and the second entropy value corresponding to the second keyword in the target space is determined based on the second feature representation. Based on the first distribution information and the first entropy value, a first separation degree corresponding to the first keyword in the target dimension is determined. The first separation degree is used to characterize the expression intensity of the first keyword in the text to be processed. The second separation degree corresponding to the second keyword under the target dimension is determined based on the second distribution information and the second entropy value. The second separation degree is used to characterize the expression intensity of the second keyword under the current evaluation text. Determine the first dimension weight corresponding to the first keyword under the target dimension, and determine the first frequency of the first keyword appearing under the first similar text cluster corresponding to the current evaluation text; The first similarity between the first keyword and the current evaluation text is determined based on the first dimension weight, the first frequency, and the first separation degree. Determine the second dimension weight corresponding to the second keyword under the target dimension, and determine the second frequency of the second keyword appearing under the second similar text cluster corresponding to the text to be processed; The second similarity between the second keyword and the text to be processed is determined based on the second dimension weight, the second frequency, and the second separation degree. The text similarity between the text to be processed and the current evaluation text is determined by fusing the first similarity and the second similarity. The first similarity is obtained according to the following formula: ； in, This represents the k-th keyword of the i-th text to be processed. With the j-th current evaluation text The first similarity between them, where count represents the number corresponding to the target dimension. This represents the weight of the first dimension corresponding to the k-th first keyword in the i-th text to be processed under the h-th target dimension. The frequency of the first keyword appearing in the first similar text cluster corresponding to the i-th text to be processed is represented by the first frequency. and Represents a constant. This represents the i-th text to be processed. The corresponding text length, This represents the average text length of the first similar text cluster corresponding to the j-th currently evaluated text.

7. The method according to claim 1, characterized in that, The step of correcting the associated rating information based on the target clustering results to obtain the target rating information corresponding to the target evaluation text includes: Obtain the relevant scoring information corresponding to each second sub-cluster in the target clustering result from the associated scoring information; Anomaly identification is performed on the relevant rating information to obtain abnormal rating information and the corresponding evaluation text; The abnormal rating information is removed from the relevant rating information to obtain the remaining rating information; The target score information is obtained by correcting the scores of the relevant evaluation texts based on the remaining score information.

8. A scoring and modification system for evaluation text, characterized in that, include: The data acquisition module is used to obtain historical consumption information corresponding to the target user's consumption under the target merchant, and to obtain the initial evaluation text of the target merchant and the initial rating information corresponding to the initial evaluation text under the historical consumption information. The sentiment analysis module is used to perform text segmentation on the initial evaluation text to obtain multiple segmented evaluation texts, and to determine a first evaluation text from the multiple segmented evaluation texts and a first remaining evaluation text after removing the first evaluation text from the multiple segmented evaluation texts; and to perform sentiment classification on the first evaluation text to obtain a first sentiment type corresponding to the first evaluation text and a first probability information corresponding to the first sentiment type. After merging any second evaluation text and the first evaluation text in the first remaining evaluation text, sentiment analysis is performed to obtain the second sentiment type and the second probability information corresponding to the second sentiment type. The first sentiment type and the second sentiment type are compared to obtain a type comparison result. When the type comparison result is that the first sentiment type and the second sentiment type are the same, a first analysis result is determined between the first evaluation text and the second evaluation text based on the first probability information and the second probability information. The first analysis result is used to characterize the analysis result of sentiment consistency between the first evaluation text and the second evaluation text. When the type comparison result is that the first sentiment type and the second sentiment type are not the same, the first analysis result between the first evaluation text and the second evaluation text is determined as a preset result. Based on the first analysis result, multiple segmented evaluation texts are merged to obtain a target merging result. Based on the target merging result, the initial evaluation text is subjected to text sentiment consistency analysis to obtain a target analysis result corresponding to the initial evaluation text. The status determination module is used to determine the target data status corresponding to the initial evaluation text based on the target analysis results and the initial scoring information. The correlation calculation module is used to calculate the information correlation degree between the historical consumption information and the initial evaluation text; The data association module is used to delete the initial evaluation text according to the target data status and the information association degree to obtain the target evaluation text, and to obtain the associated rating information corresponding to the target evaluation text from the initial rating information; The data correction module is used to perform data clustering on the target evaluation text to obtain target clustering results, and to perform data correction on the associated rating information based on the target clustering results to obtain target rating information corresponding to the target evaluation text.

9. A terminal device, characterized in that, The terminal device includes a processor and a memory; The memory is used to store computer programs; The processor is used to execute the computer program and, in executing the computer program, implement the rating modification method for the evaluation text as described in any one of claims 1 to 7.

Images