An intelligent semantic recognition method for e-commerce reviews

Abstract

The present invention relates to the field of semantic recognition, and specifically to an intelligent recognition method for semantics of e-commerce comments, including: obtaining a number of e-commerce word bag vectors and a number of sales volume data at different historical record moments; analyzing the semantic high matching of the comment content between different e-commerce word bag vectors at the same historical record moment, and screening out a number of user e-commerce word bag vectors; analyzing the support strength of different word types in the user e-commerce word bag vector for the comment content, and obtaining the user text semantic independence; obtaining the comment semantic validity according to the user e-commerce word bag vector and sales volume data at different historical record moments, and combining the user text semantic independence; according to the comment semantic validity, adaptively identifying the user's effective comment information in the user e-commerce word bag vector. The present invention improves the authenticity of user feedback data and improves the accuracy of e-commerce product evaluation.

Description

Technical Field

[0001] The present invention relates to the field of semantic recognition, and in particular to a method for intelligently recognizing the semantics of e-commerce comments. Background Art

[0002] At this stage, with the changes in time and actual environment, people will have different degrees of demand changes for the same type of products, such as fans, and comment and provide feedback on the electronic trading platform of the corresponding products. For e-commerce companies that operate products, in order to ensure that the products are as popular as possible, they need to adjust the product's operating strategy in a timely manner through user comments and feedback.

[0003] In the user's product review information, some users may use tools such as language models to make false reviews in order to achieve certain review control purposes and reduce the accuracy of e-commerce product evaluations. Therefore, it is necessary to perform semantic intelligent recognition of user review data. Summary of the invention

[0004] The present invention provides an intelligent semantic recognition method for e-commerce comments to solve the existing problem that in the user's comment information on a product, some users may use tools such as language models to make false comments, which reduces the authenticity of user feedback data and the accuracy of e-commerce's product evaluation, making it impossible for e-commerce to effectively make reasonable response strategies.

[0005] The method for intelligently identifying e-commerce review semantics of the present invention adopts the following technical solutions:

[0006] The following steps are involved:

[0007] Obtaining a number of e-commerce bag-of-words vectors and a number of sales volume data at different historical record times, wherein the e-commerce bag-of-words vectors contain multiple types of words;

[0008] According to the word types, analyze the semantic high matching of the comment content between different e-commerce word bag vectors at the same historical record time, and select several user e-commerce word bag vectors from all e-commerce word bag vectors;

[0009] Analyze the support of different word types in each user's e-commerce bag-of-words vector for the review content, and obtain the user text semantic independence of each user's e-commerce bag-of-words vector;

[0010] Based on the user e-commerce bag-of-words vectors and sales data at different historical record times, the dynamic balance between product demand and user review feedback is analyzed, and combined with the user text semantic independence, the semantic validity of the review of each user e-commerce bag-of-words vector is obtained;

[0011] According to the semantic validity of the comments, the effective comment information of the user in the user's e-commerce bag-of-words vector is adaptively identified.

[0012] Preferably, the method for obtaining the user e-commerce bag-of-words vector is:

[0013] Take any historical record moment as the target historical record moment, analyze the similarity of the semantics contained in different e-commerce word bag vectors at the target historical record moment, and obtain the semantic repetition of the e-commerce word bag vector; analyze the matching of the word order contained in different e-commerce word bag vectors at the target historical record moment, and obtain the semantic matching degree of the e-commerce word bag vector;

[0014] According to the semantic repetition and semantic matching, the machine semantic membership of the e-commerce bag-of-words vector is obtained;

[0015] According to the machine semantic membership, several user e-commerce bag-of-words vectors are selected from all e-commerce bag-of-words vectors.

[0016] Preferably, the method for obtaining the semantic repetition degree is:

[0017] The correlation between the word type distribution content between the e-commerce word bag vectors is analyzed to obtain the word type content commonality of the e-commerce word bag vectors; based on the word type content commonality, the semantic repetition of the e-commerce word bag vectors is obtained.

[0018] Preferably, the method for obtaining the semantic matching degree is:

[0019] By comparing the arrangement of word types between e-commerce bag-of-words vectors, the word order commonality of e-commerce bag-of-words vectors is obtained; based on the word order commonality, the semantic matching degree of e-commerce bag-of-words vectors is obtained.

[0020] Preferably, the method for acquiring the semantic independence of the user text is:

[0021] At any historical record time, any user e-commerce bag-of-words vector is used as the target user e-commerce bag-of-words vector, and the user e-commerce bag-of-words vectors other than the target user e-commerce bag-of-words vector are used as the reference user e-commerce bag-of-words vectors;

[0022] By comparing the richness of the word types contained in the target user's e-commerce bag-of-words vector and the e-commerce bag-of-words vectors of other reference users, the word semantic richness of the target user's e-commerce bag-of-words vector is obtained; based on the word semantic richness, the user text semantic independence of each user's e-commerce bag-of-words vector is obtained.

[0023] Preferably, the method for acquiring the semantic richness of a word is:

[0024] The importance of different words in the target user's e-commerce bag-of-words vector is analyzed to obtain the effective comment feedback of the target user's e-commerce bag-of-words vector; based on the effective comment feedback, combined with the analysis of the word content in the target user's e-commerce bag-of-words vector, the word semantic richness of the target user's e-commerce bag-of-words vector is obtained.

[0025] Preferably, after calculating the semantic independence of the user text, the method further includes:

[0026] Normalize the semantic independence of user text.

[0027] Preferably, the method for obtaining the semantic validity of the comment is:

[0028] Based on the user e-commerce bag-of-words vectors and sales data at different historical record moments, we analyze the balance between user review feedback and sales demand, and obtain the effective conversion rate of sales feedback at each historical record moment.

[0029] Combining the effective conversion rate of sales feedback with the semantic independence of user text, the semantic effectiveness of the comments of each user's e-commerce bag-of-words vector is obtained.

[0030] Preferably, the method for obtaining the effective conversion rate of sales feedback is:

[0031] Obtain the user feedback trend value and sales demand trend value at each historical record moment; analyze the changes in the user feedback trend value and sales demand trend value between the same historical record moments, and obtain the sales feedback effective conversion rate at each historical record moment.

[0032] Preferably, after calculating the semantic validity of the comment, the method further includes:

[0033] Normalize the semantic validity of comments.

[0034] The beneficial effects of the technical solution of the present invention are as follows: analyzing the semantic high matching of the comment content between different e-commerce word bag vectors at the same historical record time, screening out a number of user e-commerce word bag vectors from all e-commerce word bag vectors, and preliminarily excluding the comment information generated by language tools; analyzing the support strength of different word types in the user e-commerce word bag vector for the comment content, and obtaining the user text semantic independence of the user e-commerce word bag vector; wherein the user text semantic independence reflects the degree to which the user's individual language characteristics can be represented in the comment information of the user e-commerce word bag vector, making the user information hidden in the comment information more obvious; according to the user e-commerce word bag vector and sales volume data at different historical record times, The dynamic equilibrium state between commodity demand and user review feedback is analyzed, and the semantic validity of the user's e-commerce bag-of-words vector is obtained in combination with the user's text semantic independence; the semantic validity of the review reflects the degree to which the review information of the user's e-commerce bag-of-words vector is valuable to the subsequent business strategy of the e-commerce, and further identifies the recognition of the user's review semantics for the product in the review information; based on the analysis of the personalized features in the user's review content, the present invention further analyzes the dynamic equilibrium relationship between the semantics represented by the review content and product marketing, adaptively identifies the user's effective review information in the user's e-commerce bag-of-words vector, improves the authenticity of the user feedback data, and improves the accuracy of the e-commerce product evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0036] Figure 1 The present invention is a flowchart of the steps of the method for intelligently identifying semantics of e-commerce reviews. DETAILED DESCRIPTION

[0037] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation method, structure, features and effects of the e-commerce review semantic intelligent recognition method proposed by the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0038] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0039] The specific scheme of the method for intelligently identifying semantics of e-commerce reviews provided by the present invention is described in detail below with reference to the accompanying drawings.

[0040] See also Figure 1 , which shows a flowchart of a method for intelligently identifying semantics of e-commerce comments provided by an embodiment of the present invention, the method comprising the following steps:

[0041] Step S001: obtaining a number of e-commerce bag-of-words vectors and a number of sales volume data at different historical record times, wherein the e-commerce bag-of-words vectors contain a variety of word types.

[0042] It should be noted that in the user's review information on products, some users may use tools such as language models to make false reviews in order to achieve certain review control purposes, reduce the authenticity of user feedback data, and reduce the accuracy of e-commerce's product evaluation, resulting in e-commerce's inability to effectively make reasonable response strategies; therefore, it is necessary to perform semantic intelligent recognition of user review data.

[0043] In a specific implementation of the embodiment of the present invention, the method for obtaining the e-commerce bag-of-words vector and sales data is as follows: from all the comments on the product by all users in the e-commerce comment database in the past three months, one day is taken as a historical record moment, and a number of e-commerce bag-of-words vectors at a number of historical record moments are obtained through the bag-of-words model; and the sales data at each historical record moment are recorded. Each comment information corresponds to an e-commerce bag-of-words vector, and each e-commerce bag-of-words vector contains multiple types of words.

[0044] It should be noted that in this embodiment, the product and e-commerce type are described by taking an e-commerce company that sells fans as an example, and this embodiment does not make any specific limitations. The product and e-commerce type may be determined according to the specific implementation situation. In addition, the process of obtaining a bag-of-words vector through a bag-of-words model is a well-known technology and will not be repeated in this embodiment.

[0045] So far, several e-commerce bag-of-words vectors and several sales volume data at different historical record times are obtained through the above method.

[0046] Step S002: Analyze the semantic high matching of the comment content between different e-commerce bag-of-words vectors at the same historical record time according to the word types, and select a number of user e-commerce bag-of-words vectors from all e-commerce bag-of-words vectors.

[0047] It should be noted that for any historical record moment, if there is a user's behavior of generating comment information through language tools at that historical record moment, then the semantic content expressed by the comment information contained in these e-commerce word bag vectors will be highly matched; therefore, according to the type of words, the semantic high matching of the comment content between different e-commerce word bag vectors at the same historical record moment can be analyzed, and several user e-commerce word bag vectors can be selected from all e-commerce word bag vectors.

[0048] Preferably, in some implementations of the embodiments of the present invention, the method for obtaining the user e-commerce bag-of-words vector is as follows: taking any historical record moment as the target historical record moment, analyzing the similarity of the semantics contained in different e-commerce bag-of-words vectors at the target historical record moment, and obtaining the semantic repetition of the e-commerce bag-of-words vector; analyzing the matching of the word order contained in different e-commerce bag-of-words vectors at the target historical record moment, and obtaining the semantic matching degree of the e-commerce bag-of-words vector; obtaining the machine semantic membership of the e-commerce bag-of-words vector based on the semantic repetition degree and the semantic matching degree; and selecting a number of user e-commerce bag-of-words vectors from all e-commerce bag-of-words vectors based on the machine semantic membership degree. The specific process is as follows:

[0049] 1. Obtain semantic repetition.

[0050] It should be noted that the user's comment information is composed of multiple words to express semantics, so that the comment information as a whole expresses the user's product feedback; while the comment information automatically generated by the existing language tools usually only expresses the positive or negative feedback information of the e-commerce product as a whole; these feedback information usually contain a large number of the same word types; therefore, the correlation between the word type distribution content between the e-commerce word bag vectors can be analyzed to obtain the word type content commonality of the e-commerce word bag vector; according to the word type content commonality, the semantic repetition of the e-commerce word bag vector is obtained. Among them, if the semantic repetition is greater, it means that the word type in the corresponding e-commerce word bag vector is used more frequently in other e-commerce word bag vectors, reflecting that the semantics composed of multiple word types in the corresponding e-commerce word bag vector are more likely to overlap.

[0051] Preferably, in some implementations of the embodiments of the present invention, the method for obtaining the semantic repetition is: analyzing the correlation between the word type distribution content between the e-commerce word bag vectors to obtain the word type content universality of the e-commerce word bag vectors; and obtaining the semantic repetition of the e-commerce word bag vectors based on the word type content universality. The specific process is:

[0052] At the target historical record time, the cosine similarity between any two e-commerce word bag vectors is used as the commonality of the word category content of the two e-commerce word bag vectors; a word category content commonality threshold is preset , the commonness of the word type content is greater than Each e-commerce word bag vector is used as a high similarity e-commerce word bag vector. This example is described as an example, and this embodiment is not specifically limited. It can be determined according to the specific implementation situation; in addition, the process of obtaining the cosine similarity is a well-known technology and will not be repeated in this embodiment.

[0053] It should be noted that, for any one of all highly similar e-commerce word bag vectors, there is at least one word type whose commonality is greater than Highly similar e-commerce bag-of-words vectors.

[0054] Furthermore, taking any highly similar e-commerce bag-of-words vector as an example, the average commonality of word category content between the highly similar e-commerce bag-of-words vector and all other highly similar e-commerce bag-of-words vectors is taken as the semantic repetition of the highly similar e-commerce bag-of-words vector.

[0055] It should be noted that if the semantic repetition is greater, it means that the word types in the high-similar e-commerce word bag vector are more frequently used in other high-similar e-commerce word bag vectors, reflecting that the semantics composed of multiple word types in the e-commerce word bag vector are more likely to be reused.

[0056] 2. Obtain semantic matching degree.

[0057] It should be noted that in addition to the proportion of word types that will affect semantic expression, the semantics expressed by the same word type in different order arrangements may also have a greater impact on the semantic expression. For example, "He ignored me" and "I ignored him". The word types contained in these two sentences are exactly the same, but through different order arrangements, two obviously different semantics can be clearly expressed; therefore, the arrangement of word types between e-commerce word bag vectors can be compared to obtain the word order universality of e-commerce word bag vectors; according to the word order universality, the semantic matching degree of e-commerce word bag vectors can be obtained. Among them, if the semantic matching degree is greater, it means that the word ranking between the corresponding e-commerce word bag vector and other e-commerce word bag vectors is closer, reflecting that the semantics between the corresponding e-commerce word bag vector and other e-commerce word bag vectors are more consistent and matched.

[0058] Preferably, in some implementations of the embodiments of the present invention, the method for obtaining the semantic matching degree is: comparing the arrangement of word types between e-commerce word bag vectors to obtain the word order universality of each e-commerce word bag vector; and obtaining the semantic matching degree of each e-commerce word bag vector based on the word order universality. The specific process is:

[0059] Any high-similar e-commerce bag-of-words vector other than the high-similar e-commerce bag-of-words vector is used as a control high-similar e-commerce bag-of-words vector; each word type that is the same between the high-similar e-commerce bag-of-words vector and the control high-similar e-commerce bag-of-words vector is used as a similar word type, and the absolute value of the difference between the high-similar e-commerce bag-of-words vector and the control high-similar e-commerce bag-of-words vector when the similar word type first appears in time sequence is used as the order difference value of the similar word type.

[0060] Furthermore, as an example, the semantic matching degree can be calculated by the following formula:

[0061] ;

[0062] In the formula, Indicates the semantic matching degree of the highly similar e-commerce bag-of-words vector; Represents the number of all the comparison high similarity e-commerce bag-of-words vectors of the high similarity e-commerce bag-of-words vector; Indicates that the highly similar e-commerce word bag vector is The number of all similar word types between the highly similar e-commerce bag-of-words vectors; Indicates that the highly similar e-commerce word bag vector is Compare the highly similar e-commerce bag-of-words vectors The order difference value of similar word types.

[0063] It should be noted that if the semantic matching degree is greater, it means that the word ranking between the high-similar e-commerce word bag vector and other comparison high-similar e-commerce word bag vectors is closer, reflecting that the semantics between the high-similar e-commerce word bag vector and other comparison high-similar e-commerce word bag vectors are more consistent and matched.

[0064] 3. Obtain the user’s e-commerce bag-of-words vector.

[0065] The normalized value of the product of the semantic repetition degree and the machine semantic membership degree of the highly similar e-commerce bag-of-words vector is used as the machine semantic membership degree of the highly similar e-commerce bag-of-words vector.

[0066] It is particularly noted that this embodiment uses a linear normalization function as an example for normalization processing, wherein the normalization processing method may be determined according to specific implementation conditions.

[0067] Furthermore, a machine semantic membership threshold is preset , the machine semantic membership is greater than The highly similar e-commerce word bag vector is used as the machine e-commerce word bag vector; each e-commerce word bag vector other than the machine e-commerce word bag vector is used as the user e-commerce word bag vector. This example is described as an example, and this embodiment is not specifically limited. It may depend on the specific implementation situation.

[0068] So far, several user e-commerce bag-of-words vectors have been obtained through the above method.

[0069] Step S003: Analyze the support strength of different word types in each user's e-commerce bag-of-words vector for the review content, and obtain the user text semantic independence of each user's e-commerce bag-of-words vector.

[0070] It should be noted that for any user's e-commerce bag-of-words vector, the more word types it contains, the less content a single word type occupies in the corresponding user's comment content, which means that the single word type has less support for the overall semantics of the corresponding user's comment content, reflecting that the semantics expressed in the corresponding user's comment content has more individual independent characteristics of the user. Therefore, we can analyze the support of different word types in each user's e-commerce bag-of-words vector for the comment content and obtain the user text semantic independence of each user's e-commerce bag-of-words vector.

[0071] Preferably, in some implementations of the embodiments of the present invention, the method for obtaining the semantic independence of user text is: at any historical record time, any user e-commerce bag-of-words vector is used as the target user e-commerce bag-of-words vector, and the user e-commerce bag-of-words vectors other than the target user e-commerce bag-of-words vector are used as the reference user e-commerce bag-of-words vector; the target user e-commerce bag-of-words vector and other reference user e-commerce bag-of-words vectors are compared in terms of the richness of the types of words contained therein to obtain the word semantic richness of the target user e-commerce bag-of-words vector; according to the word semantic richness, the user text semantic independence of each user e-commerce bag-of-words vector is obtained. The specific process is as follows:

[0072] It should be noted that for any user's e-commerce bag-of-words vector, the richer the structure of the word types it contains, the more words that represent the user's individual characteristics are contained in the corresponding user's comment information, which means that the corresponding user's comment information has more reference value for the e-commerce's subsequent business strategy, reflecting that the corresponding user's comment information is more effective for the e-commerce.

[0073] Preferably, in some implementations of the embodiments of the present invention, the method for obtaining the semantic richness of words is: analyzing the importance of different words in the target user's e-commerce word bag vector to obtain the effective comment feedback of the target user's e-commerce word bag vector; according to the effective comment feedback, combined with analyzing the word content in the target user's e-commerce word bag vector, the semantic richness of the words in the target user's e-commerce word bag vector is obtained. The specific process is:

[0074] use The algorithm obtains the target user's e-commerce word bag vector for each word type The value is used as the importance of user comments for each word type; the average of the importance of user comments for all word types is used as the effective comment feedback of the target user's e-commerce bag-of-words vector. The process of value is The well-known contents of the algorithm will not be described in detail in this embodiment.

[0075] Furthermore, the number of occurrences of each word type in the target user's e-commerce bag-of-words vector is obtained, and the average number of occurrences of all word types in the target user's e-commerce bag-of-words vector is used as the average number of occurrences of word types in the target user's e-commerce bag-of-words vector; the product of the average number of occurrences of word types in the target user's e-commerce bag-of-words vector and the effective comment feedback degree is used as the word semantic richness of the target user's e-commerce bag-of-words vector.

[0076] It should be noted that if the semantic richness of the words is greater, it means that the target user's e-commerce bag-of-words vector contains more words that represent the user's individual characteristics, which means that the comment information in the target user's e-commerce bag-of-words vector has more reference value for the e-commerce's subsequent business strategy, reflecting that the corresponding user's comment information is more effective for the e-commerce.

[0077] Furthermore, the modulus of the target user's e-commerce bag-of-words vector is obtained; and the normalized value of the product of the modulus of the target user's e-commerce bag-of-words vector and the word semantic richness is used as the user text semantic independence of the target user's e-commerce bag-of-words vector.

[0078] It should be noted that if the semantic independence of user text is greater, it means that the target user's e-commerce bag-of-words vector contains more word types, which means that the content of a single word type in the corresponding user's comment content is smaller, which means that the support of a single word type in the overall semantics of the corresponding user's comment content is smaller, reflecting that the semantics expressed in the corresponding user's comment content has more individual independent characteristics of the user.

[0079] At this point, the user text semantic independence of each user's e-commerce bag-of-words vector is obtained through the above method.

[0080] Step S004: Analyze the dynamic balance between product demand and user review feedback based on the user e-commerce bag-of-words vectors and sales volume data at different historical record times, and obtain the semantic validity of the review of each user e-commerce bag-of-words vector in combination with the user text semantic independence.

[0081] It should be noted that for the same type of product, the functional requirements of users for this type of product will fluctuate in different time periods, which will in turn affect the sales of this type of product and make it change synchronously. For example, the blowing function of a fan product will have a certain cooling effect in summer, so the sales of fans will be higher in summer and lower in winter. Therefore, the dynamic balance between product demand and user comment feedback can be analyzed based on the user e-commerce bag-of-words vector and sales data at different historical records, and the semantic validity of the comments of each user's e-commerce bag-of-words vector can be obtained by combining the user text semantic independence. The greater the semantic validity of the comment, the more authentic the semantic content expressed by the corresponding user's comment information, which reflects that the corresponding user's comment information is more effective for e-commerce.

[0082] Preferably, in some implementations of the present invention, the method for obtaining the semantic validity of comments is: according to the user e-commerce word bag vector and sales volume data at different historical record moments, the balance between user comment feedback and sales demand is analyzed to obtain the effective conversion rate of sales feedback at each historical record moment; the semantic validity of the comment of each user e-commerce word bag vector is obtained by combining the effective conversion rate of sales feedback with the semantic independence of user text. The specific process is as follows:

[0083] It should be noted that as time goes by, the user's demand for products is constantly changing due to factors such as their environment, which affects the sales of the corresponding products; the more product sales, the greater the user's demand for the product, and the easier it is to use the product and provide comments and feedback to the e-commerce. Therefore, according to the user e-commerce bag-of-words vector and sales volume data at different historical record moments, the time series synchronization of user comment feedback and sales demand can be analyzed to obtain the effective conversion rate of sales feedback at each historical record moment. Among them, if the effective conversion rate of sales feedback is larger, it means that the quantity between user feedback and product sales is more balanced, reflecting that the user's feedback information on the product is more effectively converted into comment content that can be referenced by the e-commerce.

[0084] Preferably, in some implementations of the embodiments of the present invention, the method for obtaining the effective conversion rate of sales feedback is: obtaining the user feedback trend value and the sales demand trend value at each historical record moment; analyzing the changes in the user feedback trend value and the sales demand trend value at the same historical record moment, and obtaining the effective conversion rate of sales feedback at each historical record moment. The specific process is:

[0085] Taking any two adjacent historical record moments as an example, the difference in the number of user e-commerce bag-of-words vectors between the second historical record moment and the first historical record moment is taken as the user feedback trend value of the second historical record moment; the difference in sales volume data between the second historical record moment and the first historical record moment is taken as the sales demand trend value of the second historical record moment; the user feedback trend values ​​and sales demand trend values ​​of all historical record moments are obtained, and the ratio between the user feedback trend value and the sales demand trend value at each historical record moment is taken as the effective conversion rate of sales feedback at each historical record moment.

[0086] It should be noted that if the effective conversion rate of sales feedback is higher, it means that the quantity between user feedback on the product and the product sales is more balanced, reflecting that the user feedback information on the product is more effectively converted into review content that can be referenced by e-commerce.

[0087] It is particularly noted that this embodiment does not consider the user feedback trend value and the sales demand trend value at the first historical record moment.

[0088] Furthermore, the average of the user text semantic independence of all user e-commerce bag-of-words vectors at the same historical record moment is taken as the user semantic independence average; the normalized value of the product of the user semantic independence average at each historical record moment and the sales feedback effective conversion rate is taken as the comment semantic effectiveness at each historical record moment.

[0089] It should be noted that, if the semantic validity of the comment is greater, it means that the semantic content expressed by the corresponding user's comment information is more authentic, and the corresponding user's comment information is more effective for e-commerce.

[0090] At this point, the semantic validity of the comments at each historical record moment is obtained through the above method.

[0091] Step S005: Adaptively identify the user's valid comment information in the user's e-commerce bag-of-words vector according to the semantic validity of the comment.

[0092] The mean of the semantic validity of comments at all historical record moments is taken as the mean of the semantic validity of comments; the standard deviation of the semantic validity of comments at all historical record moments is obtained and used as the comment semantic validity fluctuation value; the sum of the comment semantic validity mean and the comment semantic validity fluctuation value is taken as the reasonable upper limit of comment conversion; the absolute value of the difference between the comment semantic validity mean and the comment semantic validity fluctuation value is taken as the reasonable lower limit of comment conversion; the interval formed by the reasonable upper limit of comment conversion and the reasonable lower limit of comment conversion is taken as the reasonable interval of comment semantic validity.

[0093] Furthermore, the historical record time when the comment semantic validity is within the reasonable range of comment semantic validity is regarded as the user's effective comment time; a user text semantic independence threshold is preset. , the semantic independence of user text at each user's valid comment time is greater than The user's e-commerce bag-of-words vector is used as the e-commerce review reference bag-of-words vector. This example is described as an example, and this embodiment is not specifically limited. It may depend on the specific implementation situation.

[0094] It should be noted that the comment information in the e-commerce comment reference bag-of-words vector can provide real and effective product feedback information for the subsequent business strategy of the e-commerce.

[0095] At this point, this embodiment is completed.

[0096] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for intelligent semantic recognition of e-commerce reviews, characterized in that: The method comprises the following steps: Obtaining a number of e-commerce bag-of-words vectors and a number of sales volume data at different historical record times, wherein the e-commerce bag-of-words vectors contain multiple types of words; According to the word types, analyze the semantic high matching of the comment content between different e-commerce word bag vectors at the same historical record time, and select several user e-commerce word bag vectors from all e-commerce word bag vectors; Analyze the support of different word types in each user's e-commerce bag-of-words vector for the review content, and obtain the user text semantic independence of each user's e-commerce bag-of-words vector; Based on the user e-commerce bag-of-words vectors and sales data at different historical record times, the dynamic balance between product demand and user review feedback is analyzed, and combined with the user text semantic independence, the semantic validity of the review of each user e-commerce bag-of-words vector is obtained; According to the semantic validity of the comments, adaptively identify the user's valid comment information in the user's e-commerce bag-of-words vector; The method for obtaining the semantic independence of the user text is: The TF-IDF algorithm is used to obtain the TF-IDF value of each word type in the target user's e-commerce bag-of-words vector, and used as the importance of user comment expression for each word type; the mean of the importance of user comment expression for all word types is used as the effective comment feedback of the target user's e-commerce bag-of-words vector; the product of the mean number of word type occurrences in the target user's e-commerce bag-of-words vector and the effective comment feedback is used as the word semantic richness of the target user's e-commerce bag-of-words vector; the normalized value of the product of the modulus of the target user's e-commerce bag-of-words vector and the word semantic richness is used as the user text semantic independence of the target user's e-commerce bag-of-words vector.

2. The method for intelligently identifying e-commerce review semantics according to claim 1, characterized in that: The method for obtaining the user e-commerce bag-of-words vector is: Take any historical record moment as the target historical record moment, analyze the similarity of the semantics contained in different e-commerce word bag vectors at the target historical record moment, and obtain the semantic repetition of the e-commerce word bag vector; analyze the matching of the word order contained in different e-commerce word bag vectors at the target historical record moment, and obtain the semantic matching degree of the e-commerce word bag vector; According to the semantic repetition and semantic matching, the machine semantic membership of the e-commerce bag-of-words vector is obtained; According to the machine semantic membership, several user e-commerce bag-of-words vectors are selected from all e-commerce bag-of-words vectors.

3. The method for intelligently identifying e-commerce review semantics according to claim 2 is characterized in that: The method for obtaining the semantic repetition degree is: The correlation between the word type distribution content between the e-commerce word bag vectors is analyzed to obtain the word type content commonality of the e-commerce word bag vectors; based on the word type content commonality, the semantic repetition of the e-commerce word bag vectors is obtained.

4. The method for intelligently identifying e-commerce review semantics according to claim 2 is characterized in that: The method for obtaining the semantic matching degree is: By comparing the arrangement of word types between e-commerce bag-of-words vectors, the word order commonality of e-commerce bag-of-words vectors is obtained; based on the word order commonality, the semantic matching degree of e-commerce bag-of-words vectors is obtained.

5. The method for intelligently identifying e-commerce review semantics according to claim 1, characterized in that: The method for obtaining the semantic validity of the comment is as follows: Based on the user e-commerce bag-of-words vectors and sales data at different historical record moments, we analyze the balance between user review feedback and sales demand, and obtain the effective conversion rate of sales feedback at each historical record moment. Combining the effective conversion rate of sales feedback with the semantic independence of user text, the semantic effectiveness of the comments of each user's e-commerce bag-of-words vector is obtained.

6. The method for intelligently identifying e-commerce review semantics according to claim 5 is characterized in that: The method for obtaining the effective conversion rate of sales feedback is: Obtain the user feedback trend value and sales demand trend value at each historical record moment; analyze the changes in the user feedback trend value and sales demand trend value between the same historical record moments, and obtain the sales feedback effective conversion rate at each historical record moment.

7. The method for intelligently identifying e-commerce review semantics according to claim 1, characterized in that: After calculating the semantic validity of the comment, it also includes: Normalize the semantic validity of comments.

Images