Evaluation text push method and device
By screening historical evaluation texts in the user evaluation system and using similar evaluation text generation models, combined with the term standardized review, the problems of long evaluation generation time and low accuracy in the existing technology are solved, and fast and accurate text evaluation push is achieved. Protect user privacy.
Patent Information
- Application Number
- CN201911120454.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2019-11-15
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2039-11-15
AI Technical Summary
The prior art is difficult to effectively reduce the generation time and improve the accuracy of evaluation when generating user evaluation texts. Especially when filtering historical evaluation texts based on user portraits and product information, there is a lack of efficient methods to generate texts close to the user's real evaluation.
By filtering historical evaluation texts based on user portraits and current product information, using the similar evaluation text generation model to generate corresponding evaluation texts, and performing verb specification reviews before pushing to ensure the accuracy and standardization of the generated text. This model generates source statements and target statement pairs through filtering and combining training data sets, and uses the Encoder-Decoder model to generate text, improving the diversity and accuracy of evaluation.
It effectively reduces the time for generating text evaluations, improves the accuracy and standardization of text evaluations, makes the evaluations pushed to users closer to the real evaluations, and protects user privacy.
Smart Images

Figure CN112818082B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of computer technology, specifically to the field of natural language processing technology, and in particular to a method and device for pushing evaluation texts. Background Art
[0002] In the prior art, there are two main ways for users to evaluate items or services: (1) Typical evaluation method: star rating + keyword tag + text description + picture. (2) Evaluation method based on text evaluation template: pre-configured evaluation standard template, based on user star rating, tag selection (or user portrait features), mapped to pre-defined template evaluation according to rules. Summary of the invention
[0003] The embodiments of the present application provide a method and device for pushing evaluation texts.
[0004] In a first aspect, the present application provides a method for pushing an evaluation text, the method comprising: screening the historical evaluation texts of the current product according to the user portrait and the current product information to obtain a first evaluation text; inputting the first evaluation text into a similar evaluation text generation model to obtain a second evaluation text corresponding to the first evaluation text; and pushing the second evaluation text to the user.
[0005] In some embodiments, the current product information includes the user's label selection content for the current product. According to the user portrait and the current product information, the historical evaluation text of the current product is filtered to obtain a first evaluation text, including: searching the historical evaluation text of the current product according to the keywords of the user's label selection content for the current product, and matching and filtering the search results with the user portrait to obtain the first evaluation text.
[0006] In some embodiments, the current product information includes the user's star rating information for the current product. According to the user portrait and the current product information, the historical evaluation text of the current product is filtered to obtain the first evaluation text, including: searching the historical evaluation text of the current product according to the user's star rating information for the current product, and matching and filtering the search results with the user portrait to obtain the first evaluation text.
[0007] In some embodiments, pushing the second evaluation text to the user includes: determining whether the second evaluation text meets preset terminology standards, and if so, pushing the second evaluation text to the user.
[0008] In some embodiments, the similar evaluation text generation model is trained in the following manner: an original training data set is obtained based on the user's historical evaluation information on the product; based on the number of likes and / or views, the original training data in the original training data set is screened to obtain a candidate comment data set; in the candidate comment set, the candidate comment data whose similarity exceeds a preset threshold are combined in pairs to obtain source sentence and target sentence pairs; one of the source sentence and target sentence pairs is used as the input of a pre-established text generation model, and the other of the source sentence and target sentence pairs is used as the expected output of the pre-established text generation model, and the pre-established text generation model is trained to obtain a similar evaluation text generation model.
[0009] In the second aspect, the present application provides an evaluation text pushing device, which includes: a screening module, configured to screen the historical evaluation text of the current product according to the user portrait and the current product information to obtain a first evaluation text; a generation module, configured to input the first evaluation text into a similar evaluation text generation model to obtain a second evaluation text corresponding to the first evaluation text; and a push module, configured to push the second evaluation text to the user.
[0010] In some embodiments, the current product information includes the user's label selection content for the current product, and the filtering module is further configured to: search the historical evaluation text of the current product according to the keywords of the user's label selection content for the current product, and match and filter the search results with the user portrait to obtain the first evaluation text.
[0011] In some embodiments, the current product information includes the user's star rating information for the current product, and the screening module is further configured to: search the historical evaluation text of the current product according to the user's star rating information for the current product, and match and screen the search results with the user portrait to obtain a first evaluation text.
[0012] In some embodiments, the push module is further configured to: determine whether the second evaluation text meets the preset terminology standards, and if so, push the second evaluation text to the user.
[0013] In some embodiments, a similar evaluation text generation model is trained in the following manner: an original training data set is obtained based on historical evaluation information of a product by a user; the original training data in the original training data set is screened based on the number of likes and / or views to obtain a candidate comment data set; in the candidate comment set, candidate comment data with a similarity exceeding a preset threshold are combined in pairs to obtain source sentence and target sentence pairs; one of the source sentence and target sentence pairs is used as the input of a pre-established text generation model, and the other of the source sentence and target sentence pairs is used as the expected output of the pre-established text generation model, and the pre-established text generation model is trained to obtain a similar evaluation text generation model.
[0014] In a third aspect, the present application provides an electronic device comprising one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by the one or more processors, the one or more processors implement an evaluation text push method.
[0015] In a fourth aspect, the present application provides a computer-readable medium having a computer program stored thereon, which implements an evaluation text push method when executed by a processor.
[0016] The evaluation text push method and device provided in the present application obtain a first evaluation text by screening the historical evaluation texts of the current product according to the user portrait and the current product information; input the first evaluation text into a similar evaluation text generation model to obtain a second evaluation text corresponding to the first evaluation text; and push the second evaluation text to the user, thereby effectively reducing the time for generating text evaluations and improving the accuracy of text evaluations. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;
[0018] Figure 2 is a flowchart of an embodiment of a method for pushing evaluation texts according to the present application;
[0019] Figure 3 is a schematic diagram of an application scenario of the evaluation text push method according to the present application;
[0020] Figure 4 is a flowchart of another embodiment of the evaluation text push method according to the present application;
[0021] Figure 5 is a flowchart of another embodiment of the evaluation text push method according to the present application;
[0022] Figure 6is a schematic diagram of an embodiment of an evaluation text pushing device according to the present application;
[0023] Figure 7 It is a structural diagram of a computer system suitable for implementing a server of an embodiment of the present application. DETAILED DESCRIPTION
[0024] The present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It should also be noted that, for ease of description, only the parts related to the relevant invention are shown in the accompanying drawings.
[0025] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0026] Figure 1 An exemplary system architecture 100 is shown to which an embodiment of the evaluation text push method of the present application can be applied.
[0027] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The server 105 may be a background server for providing text processing support. The network 104 is used to provide a medium for a communication link between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.
[0028] Terminal devices 101, 102, 103 can be hardware or software. When terminal devices 101, 102, 103 are hardware, they can be various electronic devices with display devices, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, etc. When terminal devices 101, 102, 103 are software, they can be installed in the electronic devices listed above. It can be implemented as multiple software or software modules (for example, software or software modules used to provide distributed services), or it can be implemented as a single software or software module. No specific limitation is made here.
[0029] The terminal devices 101, 102, and 103 are used to filter the historical evaluation texts of the current product according to the user portrait and the current product information to obtain the first evaluation text, obtain the second evaluation text through the similar evaluation text generation model, and push it to the user.
[0030] The server 105 may be a server providing various services, such as a backend server for providing review text processing support. The server 105 may process the user portrait, current product information, and historical review text of the current product, and feed back the processing result (such as the second review text) to the terminal device.
[0031] The server 105 can be hardware or software. When the server 105 is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or it can be implemented as a single server. When the server is software, it can be implemented as multiple software or software modules, or it can be implemented as a single software or software module. No specific limitation is made here.
[0032] It should be noted that the evaluation text push method provided in the embodiment of the present application can be executed by the server 105, or a part of it is executed by the terminal devices 101, 102, 103 and the other part is executed by the server 105. Correspondingly, the evaluation text push device is also mainly set in the server 105, or a part of it is set in the terminal devices 101, 102, 103 and the other part is set in the server 105.
[0033] It should be understood that Figure 1 The number of terminal devices 101, 102, 103, network 104 and server 105 is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.
[0034] Figure 2 A flowchart 200 of an embodiment of a method for pushing evaluation texts that can be applied to the present application is shown. The method for pushing evaluation texts includes the following steps:
[0035] Step 201, based on the user portrait and the current product information, the historical evaluation texts of the current product are screened to obtain a first evaluation text.
[0036] In this embodiment, the execution subject (eg Figure 1 The terminal devices 101, 102, 103 or the server 105 shown in the figure filter the historical evaluation information of the current product in a variety of filtering methods according to the user portrait and the information of the current product to obtain a first evaluation text.
[0037] Among them, user portrait refers to the label obtained by abstracting each specific information of the user. Related electronic devices can use these labels to concretize the user image, so as to provide targeted services to users. For example, in actual applications, the user portrait compiled based on user behavior can be: student / high consumption / focus on appearance and brand.
[0038] Here, the current product information may include the category information of the current product, such as mobile phones, stoves, sports shoes, etc.; the label attribute information of the current product and the likes data and pageview data corresponding to the historical evaluation text of the current product. Among them, the label attribute information is used to represent the characteristic information of different dimensions of the current product, which is mainly obtained by the platform through keyword extraction and statistical summary of the historical evaluation information of the product, such as beautiful appearance, durable battery, good quality and low price, etc.
[0039] It should be pointed out that the screening method can adopt the process of selecting substances with a certain specific property from a majority of substances according to a preset goal in existing technology or future development technology, such as keyword search, likes ranking, pageview ranking, etc. This application does not limit this.
[0040] The specific screening process may include:
[0041] (1) The executing entity first obtains the current product information and user portrait. For example, the current product information includes category = mobile phone, label attribute = beautiful appearance, durable battery, clear image, and user portrait = student / high consumption / concerned about appearance and brand.
[0042] (2) Filter the historical evaluation texts of the current mobile phone based on the obtained product information and user portrait. For example, you can first search for evaluation texts containing any one or more keywords in the tag attribute = beautiful appearance, durable battery, clear image, and obtain 100 evaluation texts that are potentially matching users. Then, further matching and screening are performed based on the user portrait, and the candidate evaluations with the most repeated features are selected as the first evaluation text. For example, all of them are students, and all of them are evaluations about the appearance and brand of the mobile phone, such as "the mobile phone has a beautiful appearance and I am very satisfied."
[0043] In addition, if there are multiple candidate reviews, the one with the highest number of likes can be selected as the first review text. If the number of likes is still the same, the one with the highest number of views can be selected as the first review text. If the number of views is still the same, a random one can be selected as the first review text.
[0044] It should be noted that if the number of historical evaluation texts of the current product is small, the historical evaluation texts of products of the same category as the current product can be screened.
[0045] Step 202, input the first evaluation text into the similar evaluation text generation model to obtain a second evaluation text corresponding to the first evaluation text. Here, when determining the similarity between texts, the vector corresponding to the text can be first obtained by word2vec (text to vector), and then the similarity between the texts can be determined by the distance between the vectors (for example, Euclidean distance or cosine distance).
[0046] In this embodiment, the execution subject inputs the first evaluation text into a similar evaluation text generation model obtained by pre-training the text generation model in various ways, and obtains a second evaluation text corresponding to the first evaluation text. The first evaluation text and the second evaluation text are similar evaluation texts, that is, the similarity between the first evaluation text and the second evaluation text is greater than or equal to a preset threshold, such as 90%.
[0047] Among them, the text generation model can adopt a model in existing technology or future development technology that is responsible for regenerating a text content based on existing text content, such as an Encoder-Decoder model, etc., and this application does not limit this.
[0048] Here, the Encoder-Decoder model can be widely used in various occasions such as machine translation, text summarization, and syntactic analysis. The Encoder-Decoder model can be regarded as a general processing model suitable for processing one sentence (or paragraph) to generate another sentence (or paragraph). For a sentence pair (X, Y) consisting of a source sentence X and a target sentence Y, given an input sentence X, the Encoder-Decoder model can be used to generate the target sentence Y. X and Y can be the same language or two different languages.
[0049] In practical applications, Encoder and Decoder generally use RNN model, which is the most commonly used deep learning model for linear sequences such as text. Improved RNN models LSTM and GRU are also frequently used models. For longer sentences, LSTM and GRU models are significantly better than RNN models. However, when the sentence length exceeds 30, the effect of LSTM model will drop sharply. Generally, Attention model will be introduced at this time. This is a neural network model that reflects the probability of alignment between output Y and input X sentence words.
[0050] In some optional methods, the similar evaluation text generation model can be trained in the following way.
[0051] First, the original training data set can be obtained based on the user's historical evaluation information on the product.
[0052] Among them, users can be multiple users or all or some users of a certain app. Products can be multiple categories of products or all or some categories of products of a certain app. Evaluation information can include star ratings, selected label content, evaluation text, etc. For example, a certain mobile phone already contains hundreds of thousands of product review information.
[0053] Next, based on the number of likes and / or page views, the original training data in the original training data set is screened to obtain a candidate comment data set.
[0054] Specifically, by setting a threshold for the number of likes and / or page views, the evaluation information with the number of likes and / or page views exceeding the preset threshold can be used to form a candidate review data set. For example, if the threshold for the number of likes and / or page views is 10,000, the evaluation information with the number of likes and / or page views exceeding 10,000 can be selected to form a candidate review data set.
[0055] Next, in the candidate comment set, candidate comment data with similarity exceeding a preset threshold are combined in pairs to obtain source sentence and target sentence pairs.
[0056] The method for calculating the similarity of the candidate review data may adopt a text similarity (distance) calculation method in the existing technology or future development technology, such as Jaccard, WMD, etc., which is not limited in this application. In addition, the similarity calculation may be assisted by the star rating and / or whether the written label items are the same and manual review.
[0057] Next, one of the source sentence and the target sentence pair is used as the input of a pre-established text generation model, and the other of the source sentence and the target sentence pair is used as the expected output of the pre-established text generation model. The pre-established text generation model is trained to obtain a similar evaluation text generation model.
[0058] Here, the source sentence and the target sentence are in an equal relationship. Each can act as a source sentence to generate a target sentence. Both are extracted and combined from the candidate comment data set, but the source sentence and the target sentence must be similar enough. For example, the source sentence is that the sports shoes are very comfortable, breathable and beautiful, and the target sentence is that the shoes are durable, comfortable to wear and breathable.
[0059] This implementation method obtains a candidate comment set by screening historical evaluation information, and combines candidate comment data in the candidate comment set whose similarity exceeds a preset threshold in pairs to obtain source sentence and target sentence pairs, and feeds them into a text generation model, trains the model, and obtains a similar evaluation text generation model, thereby improving the diversity of evaluation texts and protecting user privacy.
[0060] In addition, it should be pointed out that users can also obtain the original training data set regularly at preset intervals, such as 24 hours, and filter based on the number of likes and / or views to obtain a candidate comment data set, and then train the text generation model based on the source sentence and target sentence pairs obtained based on the candidate comment data set to obtain an updated similar evaluation text generation model to ensure the timeliness and effectiveness of the model.
[0061] Step 203: Push the second evaluation text to the user.
[0062] In this embodiment, the executing entity can push the second evaluation text generated by the similar evaluation text generation model directly to the user, or it can push it to the user after conducting corresponding standard review, such as civilized language standards, user privacy protection, etc. This application does not limit this.
[0063] Among them, the second evaluation text can be presented using a presentation device, where the presentation device can be a device that can output image information in existing technology or future development technology, such as a computer monitor, a mobile phone screen, etc., and this application does not limit this.
[0064] In addition, after receiving the second evaluation text, the user can also add new descriptions, delete inappropriate parts, polish and modify it based on the second evaluation text to provide a final text evaluation.
[0065] In some optional methods, pushing the second evaluation text to the user includes: determining whether the second evaluation text meets preset terminology standards, and if so, pushing the second evaluation text to the user.
[0066] In this implementation, the execution subject reviews the second evaluation text according to the preset terminology standards, wherein the preset terminology standards may include whether the grammatical requirements are met, whether it is a dialect description, whether it involves user privacy, whether it is a civilized term, etc. If the preset standards are met, the second evaluation text is pushed to the user, and if the preset standards are not met, it is not pushed to the user.
[0067] Specifically, if the second evaluation text is "I am 160cm tall, and these pants are a bit long", it contains the user's height information and cannot pass the review, so the execution entity will not push it to the user. If the second evaluation text is "The mobile phone battery is durable and the appearance is beautiful", it meets the terminology standards and can be pushed to the user.
[0068] This implementation method can not only improve the standardization of the terms in the output second evaluation text, but also better protect user privacy by conducting a standardized review of the terms before pushing the second evaluation text to the user.
[0069] Continue to see Figure 3 , Figure 3 It is a schematic diagram of an application scenario of the evaluation text push method according to this embodiment.
[0070] exist Figure 3In the application scenario, the execution subject 301 is based on the user portrait of the user 302 and the current product 303 information, for example, the user portrait = student / high consumption / focus on appearance and brand, the category information of the current product 303 = mobile phone, label attribute = beautiful and generous, clear picture, durable battery, and the historical evaluation text of the current product is screened by keyword retrieval, data statistics, pageview ranking, like ranking, etc. to obtain the first evaluation file 304, such as "the mobile phone battery is very durable and the screen is beautiful". The execution subject sends the first evaluation text 304 to the trained similar evaluation text generation model 305, for example, the generation model based on the Encoder-Decoder framework, and obtains the second evaluation text 306 corresponding to the first evaluation text 304, such as "the mobile phone screen is beautiful, the battery life is long, very good". The execution subject 301 pushes the second evaluation text 306 to the user. In addition, after receiving the second evaluation text, the user can also add descriptions, delete inappropriate, polish and modify based on the received second evaluation text 306 to provide a final text evaluation.
[0071] The evaluation text push method provided by the embodiment of the present disclosure obtains a first evaluation text by screening the historical evaluation texts of the current product according to the user portrait and the current product information; the first evaluation text is input into a similar evaluation text generation model to obtain a second evaluation text corresponding to the first evaluation text; and the second evaluation text is pushed to the user, which effectively reduces the time for generating text evaluation and improves the accuracy of the text evaluation.
[0072] Further references Figure 4 , which shows a process 400 of another embodiment of the evaluation text push method. In this embodiment, the product information may further include the user's tag selection content for the current product. Accordingly, the process 400 of the evaluation text push method of this embodiment may include the following steps:
[0073] Step 401, searching the historical evaluation texts of the current product according to the star rating information of the user on the current product, and matching and screening the search results with the user portrait to obtain the first evaluation text.
[0074] In this embodiment, the star rating information of the current product is obtained by the user based on the user's own favorable impression of the product usage experience. If the user has a good experience with the current product, the star rating will be higher, such as four stars or five stars. If the user has a poor experience with the current product, the star rating will be poor, such as one star or two stars.
[0075] In addition, it should be pointed out that the star rating information of the current product can be expressed in the form of stars, or in other ways that indicate the user's favorable experience of using the current product, such as favorableness, usage score, etc. This application does not limit this.
[0076] The specific screening process may include:
[0077] (1) The execution entity first obtains the user portrait of the current product and the user's star rating information for the current product. For example, the user portrait of the current product = student / high consumption / focused on appearance and brand, and the star rating = four stars.
[0078] (2) Retrieve the historical evaluation texts of the current mobile phone based on the obtained star rating information. For example, you can first retrieve the evaluation texts with a star rating of four stars. For example, after searching, you can get 100 evaluation texts that are potentially matching users. Then, further matching and screening are performed based on the user portrait, and the candidate evaluations with the most repeated features are selected and counted, such as all evaluations by students and all about the appearance and brand of the mobile phone.
[0079] In addition, if there are multiple candidate reviews, the one with the highest number of likes can be selected as the first review text. If the number of likes is still the same, the one with the highest number of views can be selected as the first review text. If the number of views is still the same, a random one can be selected as the first review text.
[0080] It should be noted that if the number of historical evaluation texts of the current product is small, the historical evaluation texts of products of the same category as the current product can be screened.
[0081] Step 402: input the first evaluation text into a similar evaluation text generation model to obtain a second evaluation text corresponding to the first evaluation text.
[0082] In this embodiment, the implementation details and technical effects of step 402 can refer to the description of step 202 and will not be repeated here.
[0083] Step 403: Push the second evaluation text to the user.
[0084] In this embodiment, the implementation details and technical effects of step 403 may refer to the description of step 203 and will not be repeated here.
[0085] This embodiment retrieves the historical evaluation information of the current product according to the user's star rating information on the current product, matches and filters the search results with the user portrait, obtains a first evaluation text, and sends the first evaluation text to the model to obtain a second evaluation text, so that the second evaluation text is closer to the user's real evaluation.
[0086] Further references Figure 5 , which shows a process 500 of another embodiment of the evaluation text push method. In this embodiment, the product information may further include the user's star rating information for the current product. Accordingly, the process 500 of the evaluation text push method of this embodiment may include the following steps:
[0087] Step 501, searching the historical evaluation texts of the current product according to the keywords of the content selected by the user for the tag of the current product, and matching and screening the search results with the user portrait to obtain the first evaluation text.
[0088] In this embodiment, the label selection content of the current product is obtained by the user's selection of the label attributes of the current product. For example, the current product is a mobile phone, and its label attributes are fast response speed, high clarity, and beautiful appearance. The user selects the label attributes according to his or her own experience of using the product, and obtains the label selection content of fast response speed.
[0089] The specific screening process may include:
[0090] (1) The execution subject first obtains the user portrait of the current product and the user's label selection content for the current product. For example, the user portrait of the current product = student / high consumption / focus on appearance and brand, and the user selects the label attributes of the current product = super fast response, durable battery, clear image based on his own usage experience, and obtains the label selection content of durable battery and clear image.
[0091] (2) Retrieve the historical evaluation texts of the current product mobile phone based on the keywords of the tag selection content obtained. For example, you can first retrieve the evaluation texts containing the keywords "battery durability" and "clear image" of the user's tag selection content for the current product. For example, after searching, you can obtain 100 evaluation texts that may potentially match the user. Then, further matching and screening are performed based on the user portrait, and the candidate evaluations with the most repeated features are selected and counted. For example, all of them are students, and all of them are about the appearance and brand of the mobile phone.
[0092] In addition, if there are multiple candidate reviews, the one with the highest number of likes can be selected as the first review text. If the number of likes is still the same, the one with the highest number of views can be selected as the first review text. If the number of views is still the same, a random one can be selected as the first review text, such as "The phone has a beautiful appearance, a durable battery, and clear images. I am very satisfied."
[0093] It should be noted that if the number of historical evaluation texts of the current product is small, the historical evaluation texts of products of the same category as the current product can be screened.
[0094] Step 502: input the first evaluation text into a similar evaluation text generation model to obtain a second evaluation text corresponding to the first evaluation text.
[0095] In this embodiment, the implementation details and technical effects of step 502 can refer to the description of step 202 and will not be repeated here.
[0096] Step 503: Push the second evaluation text to the user.
[0097] In this embodiment, the implementation details and technical effects of step 503 may refer to the description of step 203 and will not be repeated here.
[0098] This embodiment retrieves the historical evaluation texts of the current product according to the keywords of the user's tag selection content for the current product, matches and filters the search results with the user portrait, obtains the first evaluation text, and sends the first evaluation text to the model to obtain the second evaluation text, so that the second evaluation text is closer to the user's real evaluation and improves the accuracy of the evaluation.
[0099] Understandably, Figure 4 The embodiments shown and Figure 5 The embodiments shown can be combined. That is, the product information can include both the label selection content of the current product and the label selection content of the current product by the user. At this time, the execution subject can search the historical evaluation text of the current product according to the keywords of the label selection content of the current product by the user and the star rating information of the current product by the user, and obtain the search results that contain the keywords of the label selection content of the current product by the user and are the same as the star rating information of the current product by the user, and match and filter the search results with the user portrait to obtain the first evaluation text.
[0100] Further references Figure 6 As an implementation of the methods shown in the above figures, the present application provides an embodiment of an evaluation text push device, which is similar to Figure 1 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.
[0101] like Figure 6 As shown, the evaluation text push device 600 of this embodiment includes: a screening module 601, a generation module 602, and a push module 603. Among them, the screening module 601 is configured to screen the historical evaluation texts of the current product according to the user portrait and the current product information to obtain a first evaluation text; the generation module 602 is configured to input the first evaluation text into a similar evaluation text generation model to obtain a second evaluation text corresponding to the first evaluation text; the push module 603 is configured to push the second evaluation text to the user.
[0102] In this embodiment, the screening module 601 in the evaluation text pushing device 600 may use an image recognition model in the existing technology or future development technology to perform image recognition on the image to be recognized to obtain a first image.
[0103] In some optional implementations of this embodiment, the current product information includes the user's label selection content for the current product, and the filtering module 601 is further configured to: search the historical evaluation text of the current product according to the keywords of the user's label selection content for the current product, and match and filter the search results with the user portrait to obtain the first evaluation text.
[0104] In some optional implementations of this embodiment, the current product information includes the user's star rating information for the current product, and the filtering module 601 is further configured to: retrieve the historical evaluation information of the current product according to the user's star rating information for the current product, and match and filter the search results with the user portrait to obtain a first evaluation text.
[0105] In this embodiment, the screening module 601 screens the historical evaluation information of the current product using a variety of screening methods according to the user portrait and the information of the current product to obtain the first evaluation text.
[0106] In this embodiment, the generation module 602 inputs the first evaluation text into a similar evaluation text generation model obtained by pre-training the text generation model in various ways, to obtain a second evaluation text corresponding to the first evaluation text.
[0107] In some optional implementations of the present embodiment, the similar evaluation text generation model is trained in the following manner: an original training data set is obtained based on the user's historical evaluation information on the product; the original training data in the original training data set is screened based on the number of likes and / or views to obtain a candidate comment data set; in the candidate comment set, the candidate comment data whose similarity exceeds a preset threshold are combined in pairs to obtain source sentence and target sentence pairs; one of the source sentence and target sentence pairs is used as the input of a pre-established text generation model, and the other of the source sentence and target sentence pair is used as the expected output of the pre-established text generation model, and the pre-established text generation model is trained to obtain a similar evaluation text generation model.
[0108] In this embodiment, the push module 603 pushes the second evaluation text generated by the similar evaluation text generation model directly to the user, or pushes it to the user after conducting corresponding standard review, such as civilized language standards, user privacy protection, etc. This application does not limit this.
[0109] In some optional implementations of this embodiment, the push module 603 is further configured to review the second evaluation text according to preset terminology standards, wherein the preset terminology standards may include whether grammatical requirements are met, whether it is a dialect description, whether it involves user privacy, whether it is civilized language, etc. If the preset standards are met, the second evaluation text is pushed to the user, and if the preset standards are not met, it is not pushed to the user.
[0110] Reference below Figure 7 , which shows a schematic diagram of the structure of a computer system 700 suitable for implementing a client device or a server of an embodiment of the present application.
[0111] like Figure 7 As shown, the computer system 700 includes a processor (e.g., a central processing unit CPU) 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage part 708 into a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the system 600 are also stored. The CPU 601, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.
[0112] The following components are connected to the I / O interface 705: an input section 706 including a keyboard, a mouse, etc.; an output section 707 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, a modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as needed. A removable medium 711, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 710 as needed, so that a computer program read therefrom is installed into the storage section 708 as needed.
[0113] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program tangibly contained on a machine-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through the communication part 709, and / or installed from the removable medium 711.
[0114] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0115] The modules described in the embodiments of the present application may be implemented by software or by hardware. The modules described may also be set in a processor. For example, they may be described as: a processor including a screening module, a generation module, and a push module. The names of these modules do not, in some cases, constitute limitations on the modules themselves. For example, the screening module may also be described as a "module that screens the historical evaluation texts of the current product based on the user portrait and the current product information to obtain the first evaluation text."
[0116] As another aspect, the present application also provides a non-volatile computer storage medium, which can be the non-volatile computer storage medium included in the device in the above embodiment; or it can be a non-volatile computer storage medium that exists independently and is not installed in the client device. The above non-volatile computer storage medium stores one or more programs, and when the one or more programs are executed by a device, the device: screens the historical evaluation text of the current product according to the user portrait and the current product information to obtain a first evaluation text; inputs the first evaluation text into the similar evaluation text generation model to obtain a second evaluation text corresponding to the first evaluation text; and pushes the second evaluation text to the user.
[0117] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the present application.
Claims
1. A method for pushing evaluation text, the method comprising: Retrieve historical evaluation texts of the current product according to the keywords of the user's tag selection content for the current product and / or the user's star rating information for the current product, and match and filter the search results with the user portrait to obtain a first evaluation text; Inputting the first evaluation text into a similar evaluation text generation model to obtain a second evaluation text corresponding to the first evaluation text; The second evaluation text is pushed to the user, wherein the similar evaluation text generation model is trained by combining source sentences and target sentences in pairs based on candidate comment data in a candidate comment set whose similarity exceeds a preset threshold, the candidate comment set is obtained based on the user's historical evaluation information on the product, and the source sentence is different from the target sentence.
2. According to the method of claim 1, the pushing the second evaluation text to the user comprises: It is determined whether the second evaluation text meets the preset terminology standard. If so, the second evaluation text is pushed to the user.
3. According to the method according to any one of claims 1-2, the similarity evaluation text generation model is trained by the following method: According to the historical evaluation information of users on the products, the original training data set is obtained; Based on the number of likes and / or the number of views, the original training data in the original training data set is screened to obtain a candidate comment data set; In the candidate comment set, candidate comment data with similarity exceeding a preset threshold are combined in pairs to obtain source sentence and target sentence pairs; One of the source sentence and the target sentence pair is used as the input of a pre-established text generation model, and the other of the source sentence and the target sentence pair is used as the expected output of the pre-established text generation model. The pre-established text generation model is trained to obtain a similar evaluation text generation model.
4. A device for pushing evaluation texts, comprising: A screening module is configured to search the historical evaluation text of the current product according to the keywords of the content selected by the user for the tag of the current product and / or the star rating information of the user for the current product, and match and screen the search results with the user portrait to obtain a first evaluation text; A generation module configured to input the first evaluation text into a similar evaluation text generation model to obtain a second evaluation text corresponding to the first evaluation text; A push module is configured to push the second evaluation text to the user, wherein the similar evaluation text generation model is trained based on source sentences and target sentences obtained by pairing candidate comment data with similarity exceeding a preset threshold in a candidate comment set, the candidate comment set is obtained based on the user's historical evaluation information on the product, and the source sentence is different from the target sentence.
5. The device according to claim 4, wherein the push module is further configured to: It is determined whether the second evaluation text meets the preset terminology standard. If so, the second evaluation text is pushed to the user.
6. According to the device according to any one of claims 4-5, the similarity evaluation text generation model is trained by the following method: According to the historical evaluation information of users on the products, the original training data set is obtained; Based on the number of likes and / or the number of views, the original training data in the original training data set is screened to obtain a candidate comment data set; In the candidate comment set, candidate comment data with similarity exceeding a preset threshold are combined in pairs to obtain source sentence and target sentence pairs; One of the source sentence and the target sentence pair is used as the input of a pre-established text generation model, and the other of the source sentence and the target sentence pair is used as the expected output of the pre-established text generation model. The pre-established text generation model is trained to obtain a similar evaluation text generation model.
7. An electronic device comprising: one or more processors; A storage device having one or more programs stored thereon, when the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 3.
8. A computer readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method according to any one of claims 1 to 3 is implemented.
Citation Information
Patent Citations
Evaluation information generation method and device
CN104731873A
