Intelligent analysis and response system for consumer evaluation content based on artificial intelligence technology

Abstract

The present invention discloses an intelligent analysis and reply system for consumer evaluation content based on artificial intelligence technology, and belongs to the field of intelligent online shopping technology. The present invention solves the problem of insufficient practicality of the existing system, and extracts semantic features of consumer review content in various contexts in the past through a convolutional neural network model, and establishes a corresponding automatic reply template; when facing the current consumer's evaluation content, the current content is semantically analyzed to obtain semantic feature results, and the corresponding reply template is retrieved based on the result, and the consumer's evaluation content is automatically replied; by establishing a semantic feature retrieval keyword, it can quickly retrieve sample semantic features in a sample semantic feature database through the semantic features of the current consumer's comment content, and then quickly obtain an automatic reply template, thereby ensuring that the system analyzes the needs of consumer evaluation content more accurately, improving the consumer's experience, and optimizing the practicality of the automatic reply system.

Description

Technical Field

[0001] The present invention relates to the field of intelligent online shopping technology, and in particular to an intelligent analysis and reply system for consumer evaluation content based on artificial intelligence technology. Background Art

[0002] With the development of the Internet, people's shopping habits have changed a lot. Unlike the traditional way of going to shopping malls, people now shop more online. In online shopping, consumers' evaluation of products can help other consumers make purchasing decisions. However, there are many problems with traditional product evaluations, such as false evaluations, online scams, etc. Therefore, a product evaluation system based on artificial intelligence has come into being. Artificial intelligence is an intelligent system that uses computer, machine learning, natural language processing and other technologies to analyze and process massive amounts of data, thereby achieving autonomous learning and intelligent application. A product evaluation system based on artificial intelligence can automatically identify the emotional polarity, evaluation objects and main topics of the evaluation content in the evaluation, thereby providing consumers with more accurate, comprehensive and real product evaluation information.

[0003] However, the automatic reply content of existing online shopping software is mostly polite automatic reply templates, which makes it difficult to analyze the real needs of consumer evaluation content, thereby reducing the consumer experience and the practicality of the existing automatic reply system.

[0004] Therefore, the existing needs are not met, so we propose an intelligent analysis and response system for consumer evaluation content based on artificial intelligence technology. Summary of the invention

[0005] The purpose of the present invention is to provide an intelligent analysis and reply system for consumer evaluation content based on artificial intelligence technology, which extracts semantic features of consumer review content in various previous contexts through a convolutional neural network model, and establishes corresponding automatic reply templates; when facing the current consumer's evaluation content, semantic analysis is performed on the current content to obtain semantic feature results, and the corresponding reply template is retrieved based on the results, and automatically replied to the consumer's evaluation content; by establishing semantic feature retrieval keywords, it is possible to quickly retrieve sample semantic features in a sample semantic feature database through the semantic features of the current consumer's review content, and then quickly obtain an automatic reply template, thereby ensuring that the system can more accurately analyze the needs of consumer evaluation content, improve the reply efficiency of the automatic reply system, enhance the consumer's experience, optimize the practicality of the automatic reply system, and solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The intelligent analysis and response system for consumer evaluation content based on artificial intelligence technology includes:

[0008] Text transceiver terminal for

[0009] Input and receive consumer evaluation content, and obtain consumer evaluation text information;

[0010] Text processing unit for

[0011] Perform vector preprocessing on the text information input by the text transceiver terminal to obtain consumer evaluation samples for intelligent analysis, classify the evaluation samples by terms and contexts, and divide each type of evaluation samples to obtain sample training sets and sample test sets in different contexts;

[0012] Semantic analysis unit, used to

[0013] Construct a convolutional neural network model, import the sample training set in each type of context into the convolutional neural network model for training and output the sample semantic feature training results; then import the sample test set in the corresponding context into the convolutional neural network model for testing and output the sample semantic feature test results. After ensuring that the sample semantic feature training results in each type of context are consistent with the sample semantic feature test results, obtain the sample semantic feature results in different contexts to form a sample semantic feature database;

[0014] Reply establishment unit for

[0015] Obtain consumer evaluation factors based on sample semantic feature results in different terms and contexts, and make response templates in different terms and contexts for different consumer evaluation factors to form a response template database;

[0016] Automatic reply unit for

[0017] According to the semantic features of the current consumer evaluation content, a search is performed in the sample semantic feature database based on the current semantic feature information to obtain the sample semantic features that are closest to the current semantic feature information, and an automatic reply is made to the current consumer evaluation content based on the reply template corresponding to the sample semantic features.

[0018] Furthermore, the text transceiver terminal comprises

[0019] User review terminal, used to log in to the consumer evaluation system platform through wireless network technology, and assist consumers to input the evaluation content they need into the consumer evaluation system;

[0020] The user service terminal is used to exchange information with the user review terminal through wireless network technology and receive the consumer evaluation content input by the user review terminal.

[0021] Furthermore, the text processing unit includes

[0022] The text collection module is used to collect multiple groups of historical evaluation contents in different forms and contexts submitted by text sending and receiving terminals to obtain historical evaluation samples;

[0023] The text cleaning module is used to clean the historical evaluation samples, remove the evaluation content with a high proportion of typos, and then perform word vector conversion and data normalization on the historical evaluation samples.

[0024] Furthermore, the text processing unit also includes

[0025] A cleaning result real-time monitoring module is used to monitor in real time the cleaning result of each historical evaluation sample cleaning performed by the text cleaning module;

[0026] An abnormal text content acquisition module is used to extract, from the cleaning results, evaluation contents in which the proportion of wrong characters corresponding to the cleaning of each historical evaluation sample exceeds a preset proportion threshold as abnormal text contents;

[0027] An abnormal text information acquisition module is used to monitor the data volume of the abnormal text content in real time;

[0028] The abnormal impact factor acquisition module is used to obtain the abnormal impact factor of the abnormal text content by using the data volume of the abnormal text content; wherein the abnormal impact factor is obtained by the following formula:

[0029]

[0030] Among them, S represents the abnormal impact factor; n represents the number of abnormal text content collected; B i Indicates the number of typos in the i-th abnormal text content; B zi Indicates the total number of characters contained in the i-th abnormal text content; c 01 and c 02 represent the first adjustment coefficient and the second adjustment coefficient respectively;

[0031] The first adjustment coefficient is obtained by the following formula:

[0032]

[0033] Among them, c 01 represents the first adjustment coefficient; m represents the number of times the historical evaluation samples are collected; C yi represents the amount of data corresponding to the abnormal text content contained in the i-th historical evaluation sample collection; C zi Represents the total amount of text data collected from the i-th historical evaluation sample;

[0034] Furthermore, the second adjustment coefficient is obtained by the following formula:

[0035]

[0036] in, represents the second adjustment coefficient; m represents the number of times the historical evaluation samples are collected; C hi represents the cumulative amount of collected data of the historical evaluation samples corresponding to the i-th historical evaluation sample collection; C i Represents the cumulative amount of collected data corresponding to the abnormal text content after the i-th historical evaluation sample collection;

[0037] The comparison module is used to compare the abnormal impact factor with a preset factor threshold. When the abnormal impact factor exceeds the preset factor threshold, the wrong characters in the abnormal text content are corrected to obtain the correct text content corresponding to the abnormal text content.

[0038] Furthermore, the comparison module also includes

[0039] A comparison execution module, used for comparing the abnormal impact factor with a preset factor threshold to obtain a comparison result;

[0040] A data volume acquisition module, used for extracting the data volume corresponding to the information in the abnormal text when the comparison result shows that the abnormal impact factor exceeds a preset factor threshold;

[0041] A quantitative relationship extraction module, used to extract the quantitative relationship between the abnormal influencing factor and a preset factor threshold;

[0042] The text data volume range acquisition module is used to acquire the range of the text data volume that needs to be corrected for typos corresponding to the information in the abnormal text according to the abnormal impact factor and the preset factor threshold; wherein the upper limit and lower limit corresponding to the range of the text data volume are obtained by the following formula:

[0043]

[0044] Among them, C up and C down Respectively represent the upper and lower limits of the range of text data; C yz Indicates the total amount of data corresponding to the information in the abnormal text; c 01 and c 02 They represent the first adjustment coefficient and the second adjustment coefficient respectively; S represents the abnormal impact factor; S y Indicates the preset factor threshold;

[0045] A sample extraction module, used for extracting the number of historical evaluation samples in the abnormal text information that meets the range of the text data amount from the abnormal text information according to the range of the text data amount;

[0046] The correction execution module is used to correct the typos of the historical evaluation samples in the abnormal text information that meets the range of the text data volume, and obtain the correct text content corresponding to the abnormal text content.

[0047] Furthermore, the text processing unit also includes

[0048] The sample classification module is used to classify the cleaned historical evaluation samples according to the keywords of the entries and different contexts, and obtain historical evaluation samples under different entries and different contexts respectively;

[0049] The sample division unit is used to divide the data sets of historical evaluation samples under different terms and different contexts respectively, so as to obtain sample training sets and sample test sets under different terms and different contexts respectively.

[0050] Furthermore, the semantic analysis unit includes

[0051] Model building module, used to build a convolutional neural network model as a basic training model;

[0052] A semantic association module is used to associate sample training sets and sample test sets under different terms and different contexts, and obtain sample training sets and sample test sets under the same terms and the same context respectively;

[0053] The model training module is used to import the sample training set and sample test set of each same term and the same context into the basic training model in sequence, train the basic training model with the sample training set to obtain the trained convolutional neural network model, and then test the trained convolutional neural network model with the sample test set to obtain the optimized convolutional neural network model, and output the test results, i.e., the sample semantic feature results;

[0054] The semantic storage module is used to store each set of sample semantic feature results by terms and contexts to form a sample semantic feature database.

[0055] Furthermore, the semantic storage module includes

[0056] The semantic creation module is used to establish search keywords for the key words and contexts belonging to each group of sample semantic feature results.

[0057] Furthermore, the reply establishing unit includes:

[0058] The element extraction module is used to extract the semantic keywords from the semantic feature results of each group of samples and obtain the consumer evaluation elements in different terms and contexts;

[0059] The reply establishment module is used to formulate reasonable reply content templates for each group of consumer evaluation factors, and obtain reply content templates under different terms and contexts;

[0060] Common phrase creation module, used to create polite common phrase reply content templates, and transfer to manual service according to consumer evaluation requirements;

[0061] The template storage module is used to store reply content templates under different terms and contexts to form a reply template database.

[0062] Furthermore, the template storage module includes

[0063] A reply association module is used to associate multiple groups of reply content templates with multiple groups of sample semantic feature results, so that each group of reply content templates corresponds to the corresponding entry, context and its semantic feature;

[0064] The reply creation module is used to establish search keywords for the key words and contexts of each group of reply content templates.

[0065] Furthermore, the automatic reply unit comprises

[0066] A semantic retrieval module is used to retrieve the most similar sample semantic features in the sample semantic feature database according to the semantic features of the current consumer evaluation content;

[0067] A reply retrieval module is used to retrieve the corresponding reply content template in the reply template database according to the sample semantic features;

[0068] The automatic reply module is used to automatically reply the corresponding reply content template to the current consumer evaluation content through wireless network technology.

[0069] Furthermore, the automatic reply unit also includes

[0070] The transfer service module is used to obtain the consumer's transfer needs based on the semantic characteristics of the consumer's evaluation content and automatically transfer the consumer to manual service.

[0071] Compared with the prior art, the present invention has the following beneficial effects:

[0072] 1. The present invention uses a convolutional neural network model to extract semantic features of consumer review content in various contexts in the past, and establishes a corresponding automatic reply template based on this; when facing the current consumer's evaluation content, the current content is semantically analyzed to obtain semantic feature results, and the corresponding reply template is retrieved based on the results, and the reply is automatically sent to the consumer's evaluation content, thereby ensuring that the system can more accurately analyze the needs of consumer evaluation content, improve the consumer experience, and optimize the practicality of the automatic reply system.

[0073] 2. The present invention establishes semantic feature retrieval keywords so that its automatic reply system can quickly retrieve sample semantic features in the sample semantic feature database through the semantic features of the current consumer comment content, and then quickly obtain an automatic reply template, thereby improving the reply efficiency of the automatic reply system and further enhancing the consumer experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 This is a composition diagram of the consumer evaluation content intelligent analysis and response system based on artificial intelligence technology of the present invention. DETAILED DESCRIPTION

[0075] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0076] In order to solve the technical problem that the current automatic reply content of online shopping software is mostly polite automatic reply templates, which is difficult to analyze the real needs of consumer evaluation content, thereby reducing the consumer experience and the practicality of the existing automatic reply system, please refer to Figure 1 , this embodiment provides the following technical solutions:

[0077] The intelligent analysis and response system for consumer evaluation content based on artificial intelligence technology includes:

[0078] Text transceiver terminal for

[0079] Input and receive consumer evaluation content to obtain consumer evaluation text information; the text sending and receiving terminal includes:

[0080] The user review terminal is used to log in to the consumer evaluation system platform through wireless network technology, and assist consumers to input the evaluation content they need into the consumer evaluation system; specifically, the user review terminal is the user end, which can be: the user's smart phone and / or computer device, which can log in to the online shopping platform based on the wireless network, and is used to input the evaluation content into the consumer evaluation interface of the online shopping platform, so as to obtain the reply content they need.

[0081] The user service terminal is used to exchange information with the user review terminal through wireless network technology, and receive the consumer evaluation content input by the user review terminal; specifically, the user service terminal is the merchant end of the online shopping platform, which can be: a smart phone and / or computer device belonging to the merchant, which can log in to the online shopping platform based on the wireless network, receive the evaluation content input in the consumer evaluation interface of the online shopping platform, and thereby obtain the consumer's evaluation content.

[0082] In this embodiment, for example, a consumer logs in to the online shopping platform through a user review terminal and enters the evaluation page, enters the evaluation content into the page, i.e., "a certain product fades after washing", and uses this as the consumer's evaluation of the product, and waits for the system to automatically reply.

[0083] Text processing unit for

[0084] Perform vector preprocessing on the text information input by the text transceiver terminal to obtain consumer evaluation samples for intelligent analysis, classify the evaluation samples by terms and contexts, and divide each type of evaluation samples to obtain sample training sets and sample test sets in different contexts; the text processing unit includes:

[0085] The text collection module is used to collect multiple groups of historical evaluation contents in different forms and contexts submitted by text sending and receiving terminals to obtain historical evaluation samples; specifically, the text collection module is mainly used to collect previous historical evaluation records of user service terminals. The collected records need to cover a large number of different evaluation factors, thereby enriching the diversity of evaluation samples, so as to optimize the characteristic factors of the convolutional deep network model in the later stage, so that the optimized convolutional deep network model is more suitable for use in the current automatic reply system.

[0086] The text cleaning module is used to clean the historical evaluation samples, remove the evaluation content with a high proportion of typos, and then perform word vector conversion and data normalization on the historical evaluation samples; specifically, by cleaning the samples and removing invalid information, the validity of the samples is ensured; by converting the samples into word vectors, a large amount of text data in the natural language text is processed into a machine-readable vector representation for the next step of analysis and processing; since the evaluation information source is published by many consumers, the terminals used by each consumer are different, which leads to the diversity of the evaluation content data source. In order to ensure the uniformity of the sample data, the sample needs to be preprocessed to ensure the reliability of the evaluation sample.

[0087] The sample classification module is used to classify the cleaned historical evaluation samples according to the entry keywords and different contexts, and obtain historical evaluation samples under different entries and different contexts respectively; specifically, after cleaning many historical comment samples, in order to ensure the diversity of sample content, the samples cover more factors; therefore, it is necessary to classify the preprocessed comment samples, and classify the comment samples in the same context or the same entry into one category, thereby ensuring the orderliness of the comment samples, so as to facilitate the import of samples under the same context and entry into the convolutional neural network model for training and testing, thereby ensuring the diversity of features in the convolutional neural network model while ensuring the effectiveness of the convolutional neural network model training.

[0088] The sample division unit is used to divide the historical evaluation samples under different terms and different contexts into data sets, and obtain sample training sets and sample test sets under different terms and different contexts respectively; specifically, the sample training set is used to train the basic model, define the characteristics of its model, and output the training results; the sample test set is used to test the trained convolutional neural network model, output the test results, and compare the test results with the training results. If the comparison results are consistent, it indicates that the current convolutional neural network model is the optimized model; if there is an error in the comparison result, the basic model is retrained and tested to ensure the accuracy of the convolutional neural network model, so as to provide more accurate responses to consumer evaluation content in the later stage.

[0089] Specifically, the text processing unit also includes

[0090] A cleaning result real-time monitoring module is used to monitor in real time the cleaning result of each historical evaluation sample cleaning performed by the text cleaning module;

[0091] An abnormal text content acquisition module is used to extract, from the cleaning results, evaluation contents in which the proportion of wrong characters corresponding to the cleaning of each historical evaluation sample exceeds a preset proportion threshold as abnormal text contents;

[0092] An abnormal text information acquisition module is used to monitor the data volume of the abnormal text content in real time;

[0093] The abnormal impact factor acquisition module is used to obtain the abnormal impact factor of the abnormal text content by using the data volume of the abnormal text content; wherein the abnormal impact factor is obtained by the following formula:

[0094]

[0095] Among them, S represents the abnormal impact factor; n represents the number of abnormal text content collected; B i Indicates the number of typos in the i-th abnormal text content; B ziIndicates the total number of characters contained in the i-th abnormal text content; c 01 and c 02 represent the first adjustment coefficient and the second adjustment coefficient respectively;

[0096] The first adjustment coefficient is obtained by the following formula:

[0097]

[0098] Among them, c 01 represents the first adjustment coefficient; m represents the number of times the historical evaluation samples are collected; C yi represents the amount of data corresponding to the abnormal text content contained in the i-th historical evaluation sample collection; C zi Represents the total amount of text data collected from the i-th historical evaluation sample;

[0099] Furthermore, the second adjustment coefficient is obtained by the following formula:

[0100]

[0101] in, represents the second adjustment coefficient; m represents the number of times the historical evaluation samples are collected; C hi represents the cumulative amount of collected data of the historical evaluation samples corresponding to the i-th historical evaluation sample collection; C i Represents the cumulative amount of collected data corresponding to the abnormal text content after the i-th historical evaluation sample collection;

[0102] The comparison module is used to compare the abnormal impact factor with a preset factor threshold. When the abnormal impact factor exceeds the preset factor threshold, the wrong characters in the abnormal text content are corrected to obtain the correct text content corresponding to the abnormal text content.

[0103] The technical effect of the above technical solution is: through the real-time monitoring module of cleaning results, the cleaning effect of the text cleaning module when processing historical evaluation samples can be tracked and monitored in real time. This real-time monitoring mechanism helps to timely discover possible problems or anomalies in the cleaning process, thereby improving the accuracy and efficiency of data processing.

[0104] The abnormal text content acquisition module can automatically extract the evaluation content containing a large number of typos from the cleaning results as abnormal text content. This screening method based on the percentage of typos exceeding the preset threshold can accurately identify text content that requires special attention and provide strong support for subsequent processing.

[0105] The abnormal impact factor acquisition module uses the above calculation formula to obtain the abnormal impact factor by comprehensively considering the data volume, wrong character ratio and historical data collection of the abnormal text content. This mechanism can dynamically evaluate the impact of abnormal text on the overall data quality and provide a scientific basis for subsequent decision-making and processing.

[0106] The introduction of the first adjustment coefficient and the second adjustment coefficient makes the calculation of the abnormal impact factor more flexible and accurate. The two adjustment coefficients are adjusted according to the number of collection times of historical evaluation samples and the cumulative collection data volume of abnormal text content, respectively, which can adaptively reflect the trend and characteristics of data changes and improve the accuracy and effectiveness of the evaluation.

[0107] When the abnormal impact factor exceeds the preset factor threshold, the comparison module triggers the correction mechanism to correct the typos in the abnormal text content. This intelligent correction function can automatically fix errors in the data, improve the accuracy and availability of the data, and provide a reliable data foundation for subsequent text analysis and decision support.

[0108] In summary, this technical solution effectively improves the ability to identify and process abnormal text content during text processing through a series of technical means such as real-time monitoring, precise identification, dynamic evaluation and intelligent correction, improves the accuracy and efficiency of data processing, and provides strong guarantees for subsequent text analysis and decision support.

[0109] Specifically, the comparison module also includes

[0110] A comparison execution module, used for comparing the abnormal impact factor with a preset factor threshold to obtain a comparison result;

[0111] A data volume acquisition module, used for extracting the data volume corresponding to the information in the abnormal text when the comparison result shows that the abnormal impact factor exceeds a preset factor threshold;

[0112] A quantitative relationship extraction module, used to extract the quantitative relationship between the abnormal influencing factor and a preset factor threshold;

[0113] The text data volume range acquisition module is used to acquire the range of the text data volume that needs to be corrected for typos corresponding to the information in the abnormal text according to the abnormal impact factor and the preset factor threshold; wherein the upper limit and lower limit corresponding to the range of the text data volume are obtained by the following formula:

[0114]

[0115] Among them, C up and C down Respectively represent the upper and lower limits of the range of text data; Cyz Indicates the total amount of data corresponding to the information in the abnormal text; c 01 and c 02 They represent the first adjustment coefficient and the second adjustment coefficient respectively; S represents the abnormal impact factor; S y Indicates the preset factor threshold;

[0116] A sample extraction module, used for extracting the number of historical evaluation samples in the abnormal text information that meets the range of the text data amount from the abnormal text information according to the range of the text data amount;

[0117] The correction execution module is used to correct the typos of the historical evaluation samples in the abnormal text information that meets the range of the text data volume, and obtain the correct text content corresponding to the abnormal text content.

[0118] The technical effect of the above technical solution is: by introducing a text data volume range acquisition module, the technical solution can dynamically calculate the range of text data volume that needs to be corrected for typos (i.e., the upper and lower limits) based on the quantitative relationship between the abnormal impact factor and the preset factor threshold. This refined control mechanism helps reduce unnecessary correction work while ensuring that key data is corrected in a timely and accurate manner.

[0119] Through the application of the first adjustment coefficient and the second adjustment coefficient, the range calculation of the text data volume can take into account the number of times the historical evaluation samples are collected, the cumulative data volume of abnormal text content, and the changes in abnormal influencing factors. This intelligent adjustment mechanism enables the correction intensity to be dynamically adjusted according to the actual situation, thereby optimizing processing efficiency while ensuring data quality.

[0120] The sample extraction module can accurately extract historical evaluation samples that meet the range of the calculated text data volume from the information in the abnormal text. This precise extraction method helps reduce errors and omissions in the processing process and improve the accuracy and effectiveness of the correction work.

[0121] The correction execution module corrects the typos in the extracted historical evaluation samples and obtains the correct text content corresponding to the abnormal text content. This process realizes the rapid and accurate correction of abnormal text content, helps to improve the overall data quality, and provides a reliable data foundation for subsequent data analysis and decision support.

[0122] The design of the entire comparison module realizes an automated process from comparing abnormal impact factors to correcting typos. This highly automated processing method not only reduces the need for manual intervention and the risk of human error, but also improves the speed and efficiency of data processing.

[0123] In summary, this technical solution effectively improves the ability to recognize and process abnormal text content during text processing by introducing mechanisms such as text data range acquisition, intelligent adjustment of correction intensity, precise sample extraction, and efficient typo correction, thereby improving the accuracy and efficiency of data processing and providing strong guarantees for subsequent text analysis and decision support.

[0124] Semantic analysis unit, used to

[0125] Construct a convolutional neural network model, import the sample training set in each context into the convolutional neural network model for training and output the sample semantic feature training results; then import the sample test set in the corresponding context into the convolutional neural network model for testing and output the sample semantic feature test results. After ensuring that the sample semantic feature training results in each context are consistent with the sample semantic feature test results, obtain the sample semantic feature results in different contexts to form a sample semantic feature database; the semantic analysis unit includes:

[0126] The model building module is used to construct a convolutional neural network model as a basic training model; specifically, as can be seen from the above, the basic training model forms an optimized convolutional neural network model after training and testing with a sample training set and a sample test set.

[0127] The semantic association module is used to associate sample training sets and sample test sets under different terms and different contexts, and obtain sample training sets and sample test sets under each same term and the same context respectively; specifically, after dividing the historical evaluation samples into training sets and test sets, this part continues to divide the training sets and test sets belonging to the same term and context into a group according to the terms and contexts, so as to ensure the uniformity of the factors belonging to the sample set when optimizing the basic training model.

[0128] The model training module is used to import the sample training set and sample test set of each same term and the same context into the basic training model in sequence, train the basic training model with the sample training set to obtain the trained convolutional neural network model, and then test the trained convolutional neural network model with the sample test set to obtain the optimized convolutional neural network model, and output the test results, i.e., the sample semantic feature results;

[0129] Continuing with the above embodiment, for example, a group of sample sets is selected, and the evaluation content factor of this group is: "A certain product fades after washing". After it is imported into the convolutional neural network model for training and testing, the output test result is: "The product fades after washing". Therefore, the sample semantic feature result of the evaluation of this group of samples is: "The product fades after washing".

[0130] The semantic storage module is used to store each group of sample semantic feature results according to terms and contexts to form a sample semantic feature database; specifically, after optimizing multiple groups of sample sets with different terms and contexts through a convolutional neural network model, different sample semantic feature results can be output; therefore, in order to improve the orderliness of the sample semantic feature results, each group of sample semantic feature results can be divided according to the context and term to which it belongs.

[0131] Semantic storage module, including:

[0132] The semantic creation module is used to establish search keywords for the key words and contexts belonging to each group of sample semantic feature results; specifically, by establishing relevant term search keywords, it is mainly convenient for the user service terminal to quickly retrieve the corresponding sample semantic feature results in the sample semantic feature database according to the semantic features of the current consumer evaluation content, thereby improving the efficiency and accuracy of automatic replies.

[0133] Continuing with the above embodiment, for example, after testing with a convolutional neural network model, the sample semantic feature result of a group of sample evaluations is obtained as: "product washing and bleaching". According to this result, "product washing", "washing and bleaching", and "bleaching" can be used as search keywords under this semantic feature.

[0134] Reply establishment unit for

[0135] Consumer evaluation factors are obtained based on the sample semantic feature results under different terms and contexts, and reply templates under different terms and contexts are made for different consumer evaluation factors to form a reply template database; specifically, the construction of the reply template database is the main architecture of the automatic reply system, which collects a large number of representative historical samples as the basis of questions and makes reasonable replies on this basis, laying the foundation for subsequent automatic replies to current consumers.

[0136] Reply to establishment unit, including:

[0137] The element extraction module is used to extract the semantic keywords from the semantic feature results of each group of samples and obtain the consumer evaluation elements in different terms and contexts;

[0138] Continuing with the above embodiment, for example, a consumer currently logs in to the online shopping platform through a user review terminal and enters the evaluation page, and inputs evaluation content to the page, namely: "A certain product fades after washing". After this segment is imported into the convolutional neural network model for testing, the test result is output as: "The product fades after washing", which is the semantic feature result, namely: "The product fades after washing", thereby obtaining the consumer evaluation factor.

[0139] The reply establishment module is used to formulate reasonable reply content templates for each group of consumer evaluation factors, and obtain reply content templates under different terms and contexts;

[0140] Continuing with the above embodiment, for example, if the historical evaluation content input by consumers is collected as: "the certain product fades after washing", after being tested by the convolutional neural network model, the sample semantic feature result is obtained as: "the product fades after washing"; the reply establishment module can use "the product fades after washing" as the main problem factor to reply with a rational explanation for this problem, such as: "this product is a denim product. During the dyeing process of denim, it is affected by factors such as dyeing time and dye quality. Discoloration will occur after washing in the initial stage. This is a normal phenomenon"; and provide consumers with rational suggestions, such as: "When washing the product, choose warm water for washing, and control the water temperature at about 30°C. You can also pour a small amount of food additives into the warm water to soak the product to reduce the occurrence of discoloration."

[0141] The common phrase establishment module is used to establish polite common phrase reply content templates, and transfer to manual service according to consumer evaluation needs; specifically, the establishment of common phrase content templates is mainly to prevent the current consumer from giving an evaluation content. After analysis, it is found that the sample semantic feature database and the reply template database have not retrieved the relevant sample semantics and reply templates, so that the existing reply template database cannot solve the current consumer's problem. In this case, a polite reply can be issued to the user end first, and then the user end is transferred to manual service for resolution; or when the consumer requests manual service by himself, a polite reply can be issued to the user end first, and then the user end is transferred to manual service.

[0142] The template storage module is used to store the reply content templates under different terms and contexts to form a reply template database; the template storage module includes:

[0143] The reply association module is used to associate multiple groups of reply content templates with multiple groups of sample semantic feature results, so that each group of reply content templates corresponds to the corresponding terms, contexts and semantic features; specifically, by formulating different reply templates for different sample semantic feature results, in order to improve the orderliness of the reply templates, the sample semantic feature results of the context, terms and corresponding attributes to which each group of reply templates belongs can be associated, so that when searching according to the semantic features of the current consumer evaluation content in the later stage, the corresponding reply template can be retrieved simultaneously with the corresponding sample semantic feature results, thereby quickly replying to the current consumer evaluation content, thereby improving the efficiency of the automatic reply system.

[0144] The reply creation module is used to establish search keywords for the key words and contexts of each group of reply content templates, mainly to improve the efficiency and accuracy of automatic replies.

[0145] Automatic reply unit for

[0146] According to the semantic features of the current consumer evaluation content, a search is performed in a sample semantic feature database based on the current semantic feature information to obtain a sample semantic feature that is closest to the current semantic feature information, and an automatic reply is performed to the current consumer evaluation content based on a reply template corresponding to the sample semantic feature; the automatic reply unit includes:

[0147] A semantic retrieval module is used to retrieve the most similar sample semantic features in the sample semantic feature database according to the semantic features of the current consumer evaluation content;

[0148] Continuing with the above embodiment, for example, a consumer currently logs in to an online shopping platform through a user review terminal and enters an evaluation page, and inputs evaluation content into the page, namely, "a certain product becomes discolored after washing"; after testing through a convolutional neural network model, a sample semantic feature result is obtained as follows: "the product becomes discolored after washing"; by searching the result in a sample semantic feature database, a corresponding sample semantic feature is obtained, namely, "a certain product becomes discolored after washing" in the previous historical sample semantic features.

[0149] A reply retrieval module is used to retrieve the corresponding reply content template in the reply template database according to the sample semantic features;

[0150] Continuing with the above embodiment, since the sample semantic feature results have been associated with the corresponding reply templates, such as: when the sample semantic feature "a certain product discolors after washing" is retrieved, the corresponding reply template can be quickly obtained based on the result, namely: "This product is a denim product. During the dyeing process of denim, it is affected by factors such as dyeing time and dye quality. Discoloration will occur after the initial washing, which is a normal phenomenon. When washing the product, choose warm water for washing, and the water temperature should be controlled at about 30°C. You can also pour a small amount of food additives into the warm water to soak the product to reduce the occurrence of discoloration."

[0151] The automatic reply module is used to automatically reply the corresponding reply content template to the current consumer evaluation content through wireless network technology;

[0152] Continuing with the above embodiment, for example, the above rational explanations and suggestions can be automatically restored to the content of the current consumer evaluation, that is, "This product is a denim product. During the dyeing process of denim, it is affected by factors such as dyeing time and dye quality. Discoloration will occur after washing in the initial stage. This is a normal phenomenon. When washing the product, choose warm water for washing, and the water temperature should be controlled at about 30°C. A small amount of food can also be poured into the warm water to soak the product to reduce the occurrence of discoloration."

[0153] The transfer service module is used to obtain the consumer's transfer needs based on the semantic characteristics of the consumer's evaluation content and automatically transfer the consumer to manual service; specifically, the transfer service module corresponds to the above-mentioned common phrase establishment module, and is mainly used to first send a polite reply to the user end when the consumer requests manual service by himself, and then transfer the user end to manual service.

[0154] The beneficial effects achieved by the above content: Through the above method, the automatic reply system can more accurately analyze the needs of consumers for evaluation content, improve the reply efficiency of the automatic reply system, enhance the consumer experience, and optimize the practicality of the automatic reply system.

[0155] Working principle: The convolutional neural network model is used to extract semantic features of consumer comments in various contexts in the past, and the corresponding automatic reply template is established based on this; when facing the current consumer's evaluation content, the current content is semantically analyzed to obtain semantic feature results, and the corresponding reply template is retrieved based on the results, and automatically replied to the consumer's evaluation content; secondly, by establishing semantic feature retrieval keywords, the automatic reply system can quickly retrieve sample semantic features in the sample semantic feature database through the semantic features of the current consumer's comment content, and then quickly obtain the automatic reply template, thereby improving the reply efficiency of the automatic reply system.

[0156] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0157] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and alterations may be made to the embodiments without departing from the principles and spirit of the invention.

Claims

1. Intelligent analysis and response system for consumer evaluation content based on artificial intelligence technology, including: Text transceiver terminal for Input and receive consumer evaluation content, and obtain consumer evaluation text information; Text processing unit for Perform vector preprocessing on the text information input by the text transceiver terminal to obtain consumer evaluation samples for intelligent analysis, classify the evaluation samples by terms and contexts, and divide each type of evaluation samples to obtain sample training sets and sample test sets in different contexts; Semantic analysis unit, used to Construct a convolutional neural network model, import the sample training set in each context into the convolutional neural network model for training and output the sample semantic feature training results; Then, the sample test set in the corresponding context is put into the convolutional neural network model for testing and the sample semantic feature test results are output. After ensuring that the sample semantic feature training results in each type of context are consistent with the sample semantic feature test results, the sample semantic feature results in different contexts are obtained to form a sample semantic feature database; Reply establishment unit for Obtain consumer evaluation factors based on sample semantic feature results in different terms and contexts, and make response templates in different terms and contexts for different consumer evaluation factors to form a response template database; Automatic reply unit for According to the semantic features of the current consumer evaluation content, a search is performed in the sample semantic feature database based on the current semantic feature information to obtain the sample semantic features that are closest to the current semantic feature information, and an automatic reply is made to the current consumer evaluation content based on the reply template corresponding to the sample semantic features.

2. The intelligent analysis and reply system for consumer evaluation content based on artificial intelligence technology according to claim 1 is characterized by: The text transceiver terminal comprises User review terminal, used to log in to the consumer evaluation system platform through wireless network technology, and assist consumers to input the evaluation content they need into the consumer evaluation system; The user service terminal is used to exchange information with the user review terminal through wireless network technology and receive the consumer evaluation content input by the user review terminal.

3. The intelligent analysis and reply system for consumer evaluation content based on artificial intelligence technology according to claim 1 is characterized by: The text processing unit includes The text collection module is used to collect multiple groups of historical evaluation contents in different forms and contexts submitted by text sending and receiving terminals to obtain historical evaluation samples; The text cleaning module is used to clean the historical evaluation samples, remove the evaluation content with a high proportion of typos, and then perform word vector conversion and data normalization on the historical evaluation samples.

4. The intelligent analysis and reply system for consumer evaluation content based on artificial intelligence technology according to claim 3 is characterized by: The text processing unit also includes A cleaning result real-time monitoring module is used to monitor in real time the cleaning result of each historical evaluation sample cleaning performed by the text cleaning module; An abnormal text content acquisition module is used to extract, from the cleaning results, evaluation contents in which the proportion of wrong characters corresponding to the cleaning of each historical evaluation sample exceeds a preset proportion threshold as abnormal text contents; An abnormal text information acquisition module is used to monitor the data volume of the abnormal text content in real time; The abnormal impact factor acquisition module is used to obtain the abnormal impact factor of the abnormal text content by using the data volume of the abnormal text content; wherein the abnormal impact factor is obtained by the following formula: Among them, S represents the abnormal impact factor; n represents the number of abnormal text content collected; B i Indicates the number of typos in the i-th abnormal text content; B zi Indicates the total number of characters contained in the i-th abnormal text content; c 01 and c 02 represent the first adjustment coefficient and the second adjustment coefficient respectively; The first adjustment coefficient is obtained by the following formula: Among them, c 01 represents the first adjustment coefficient; m represents the number of times the historical evaluation samples are collected; C yi represents the amount of data corresponding to the abnormal text content contained in the i-th historical evaluation sample collection; C zi Represents the total amount of text data collected from the i-th historical evaluation sample; Furthermore, the second adjustment coefficient is obtained by the following formula: in, represents the second adjustment coefficient; m represents the number of times the historical evaluation samples are collected; C hi represents the cumulative amount of collected data of the historical evaluation samples corresponding to the i-th historical evaluation sample collection; C i Represents the cumulative amount of collected data corresponding to the abnormal text content after the i-th historical evaluation sample collection; The comparison module is used to compare the abnormal impact factor with a preset factor threshold. When the abnormal impact factor exceeds the preset factor threshold, the wrong characters in the abnormal text content are corrected to obtain the correct text content corresponding to the abnormal text content.

5. The intelligent analysis and reply system for consumer evaluation content based on artificial intelligence technology according to claim 4 is characterized by: The comparison module also includes A comparison execution module, used for comparing the abnormal impact factor with a preset factor threshold to obtain a comparison result; A data volume acquisition module, used for extracting the data volume corresponding to the information in the abnormal text when the comparison result shows that the abnormal impact factor exceeds a preset factor threshold; A quantitative relationship extraction module, used to extract the quantitative relationship between the abnormal influencing factor and a preset factor threshold; The text data volume range acquisition module is used to acquire the range of the text data volume that needs to be corrected for typos corresponding to the information in the abnormal text according to the abnormal impact factor and the preset factor threshold; wherein the upper limit and lower limit corresponding to the range of the text data volume are obtained by the following formula: Among them, C up and C down Respectively represent the upper and lower limits of the range of text data; C yz Indicates the total amount of data corresponding to the information in the abnormal text; c 01 and c 02 They represent the first adjustment coefficient and the second adjustment coefficient respectively; S represents the abnormal impact factor; S y Indicates the preset factor threshold; A sample extraction module, used for extracting the number of historical evaluation samples in the abnormal text information that meets the range of the text data amount from the abnormal text information according to the range of the text data amount; The correction execution module is used to correct the typos of the historical evaluation samples in the abnormal text information that meets the range of the text data volume, and obtain the correct text content corresponding to the abnormal text content.

6. The intelligent analysis and reply system for consumer evaluation content based on artificial intelligence technology according to claim 3 is characterized by: The text processing unit also includes The sample classification module is used to classify the cleaned historical evaluation samples according to the keywords of the entries and different contexts, and obtain historical evaluation samples under different entries and different contexts respectively; The sample division unit is used to divide the data sets of historical evaluation samples under different terms and different contexts respectively, so as to obtain sample training sets and sample test sets under different terms and different contexts respectively.

7. The intelligent analysis and reply system for consumer evaluation content based on artificial intelligence technology according to claim 1 is characterized by: The semantic analysis unit includes Model building module, used to build a convolutional neural network model as a basic training model; A semantic association module is used to associate sample training sets and sample test sets under different terms and different contexts, and obtain sample training sets and sample test sets under the same terms and the same context respectively; The model training module is used to import the sample training set and sample test set of each same term and the same context into the basic training model in sequence, train the basic training model with the sample training set to obtain the trained convolutional neural network model, and then test the trained convolutional neural network model with the sample test set to obtain the optimized convolutional neural network model, and output the test results, i.e., the sample semantic feature results; The semantic storage module is used to store each set of sample semantic feature results by terms and contexts to form a sample semantic feature database.

8. The intelligent analysis and reply system for consumer evaluation content based on artificial intelligence technology according to claim 7 is characterized by: The semantic storage module includes The semantic creation module is used to establish search keywords for the key words and contexts belonging to each group of sample semantic feature results.

9. The intelligent analysis and reply system for consumer evaluation content based on artificial intelligence technology according to claim 1 is characterized by: The reply establishing unit comprises The element extraction module is used to extract the semantic keywords from the semantic feature results of each group of samples and obtain the consumer evaluation elements in different terms and contexts; The reply establishment module is used to formulate reasonable reply content templates for each group of consumer evaluation factors, and obtain reply content templates under different terms and contexts; Common phrase creation module, used to create polite common phrase reply content templates, and transfer to manual service according to consumer evaluation requirements; The template storage module is used to store reply content templates under different terms and contexts to form a reply template database.

10. The intelligent analysis and reply system for consumer evaluation content based on artificial intelligence technology according to claim 9 is characterized by: The template storage module includes A reply association module is used to associate multiple groups of reply content templates with multiple groups of sample semantic feature results, so that each group of reply content templates corresponds to the corresponding entry, context and its semantic feature; The reply creation module is used to establish search keywords for the key words and contexts of each group of reply content templates.

11. The intelligent analysis and reply system for consumer evaluation content based on artificial intelligence technology according to claim 1 is characterized by: The automatic reply unit comprises A semantic retrieval module is used to retrieve the most similar sample semantic features in the sample semantic feature database according to the semantic features of the current consumer evaluation content; A reply retrieval module is used to retrieve the corresponding reply content template in the reply template database according to the sample semantic features; The automatic reply module is used to automatically reply the corresponding reply content template to the current consumer evaluation content through wireless network technology.

12. The intelligent analysis and reply system for consumer evaluation content based on artificial intelligence technology according to claim 11 is characterized by: The automatic reply unit also includes The transfer service module is used to obtain the consumer's transfer needs based on the semantic characteristics of the consumer's evaluation content and automatically transfer the consumer to manual service.

Images