Method, device, equipment and storage medium for generating comments

Through the method based on content analysis model and probability analysis, comments are generated, which solves the problems of low efficiency of manual comments and lack of attractiveness of template automatic comments, and achieves efficient and diverse comment content and enhances user interaction.

CN114519352BActive Publication Date: 2024-11-22BEIJING SANKUAI ONLINE TECH CO LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202210158264.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-02-21
Publication Date
2024-11-22
Estimated Expiration
2042-02-21

AI Technical Summary

Technical Problem

In the prior art, manual comments are inefficient, template-style automatic comments are unattractive, and they cannot interact with users effectively.

Method used

By determining multiple content representation words for information to be commented based on the content analysis model, combining the probability analysis of core words and comment words, the selected words in the comments are determined, and the comments are generated.

Benefits of technology

It improves comment efficiency, reduces manpower investment, increases the diversity and attractiveness of comment content, and enhances interaction with users.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114519352B_ABST
    Figure CN114519352B_ABST
Patent Text Reader

Abstract

The present application discloses a method, device, equipment and storage medium for generating comments, and belongs to the field of Internet technology. The method comprises: first determining a plurality of content representation words corresponding to the information to be commented on; then, determining the core words and the probability of each core word being located in the first position of the comment based on the plurality of content representation words, and at the same time, determining the comment words and the probability of each comment word being located in the first position of the comment; then, determining the selected words in the first position of the comment, and based on the selected words in the first position and the content representation words, the selected words in other positions of the comment; finally, based on the selected words in each position, determining the comment corresponding to the information to be commented on. The method improves the efficiency of comments, reduces manpower input, increases the diversity of comment sentences, and enhances the attractiveness of the comment content to users, thereby facilitating interaction with users.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and in particular to a method, device, equipment and storage medium for generating comments. Background Art

[0002] With the advent of the mobile Internet era, people will share social news, personal life and consumption experience in various applications (such as blogs, books, groups, etc.). Application developers can comment on the content posted by users and interact with users to increase user stickiness to the application.

[0003] Common comment methods include manual comment and template-based automatic comment. Manual comment refers to the comment made by the staff of the application after reading the information to be commented (i.e. the content posted by the user). When the template-based automatic comment method is adopted, the server corresponding to the application can pre-store multiple template-based comment words, and the server can determine the corresponding target template-based comment words according to the information to be commented, and use them to comment on the information to be commented.

[0004] Of the two commenting methods mentioned above, manual commenting is relatively inefficient and requires a lot of manpower for applications with a large number of users. Template-based automatic commenting has a limited number of templates, a single template sentence structure, and boring comment content, which makes the comment content unattractive and unable to form interaction with users. Summary of the invention

[0005] The embodiments of the present application provide a method, apparatus, device and storage medium for generating comments, which can solve the problems of low efficiency of manual comments and lack of attractiveness of template-based automatic comments in the prior art. The technical solution is as follows:

[0006] In a first aspect, a method for generating a comment is provided, the method comprising:

[0007] Determine multiple content representation words corresponding to the information to be reviewed based on the content analysis model;

[0008] Based on the core word analysis model, determining the core words in the plurality of content representation words and the probability that each core word is located in the first position in the comment of the information to be commented on;

[0009] Determine, based on a comment word determination model and the plurality of content representation words, a comment word corresponding to the information to be commented on and a probability that each comment word is located at the first position;

[0010] Determine a selected word at the first position in the comment based on the core word, the probability that each core word is located at the first position, the comment word, and the probability that each comment word is located at the first position;

[0011] Based on the selected word at the first position and the multiple content representation words, the selected words at other positions in the comment are determined, and the selected words at the first position and the selected words at other positions are combined to obtain the comment.

[0012] In a possible implementation, after determining the plurality of content representation words corresponding to the information to be reviewed based on the content analysis model, the method further includes:

[0013] Based on the probability analysis model, determining the probability that each content representation word is located at the first position;

[0014] The determining, based on the core word analysis model, the core words in the plurality of content representation words and the probability that each core word is located at the first position in the comment of the information to be commented on comprises:

[0015] Based on the core word analysis model, the plurality of content representation words and the probability of each content representation word being located at the first position, a core word in the plurality of content representation words and the probability of each core word being located at the first position are determined.

[0016] In a possible implementation, determining the comment words corresponding to the information to be commented on and the probability of each comment word being located at the first position based on the comment word determination model and the plurality of content representation words includes:

[0017] Based on the comment word determination model, the plurality of content representation words and the probability of each content representation word being located at the first position, the comment words corresponding to the information to be commented on and the probability of each comment word being located at the first position are determined.

[0018] In a possible implementation, after determining the comment word corresponding to the information to be commented on and the probability of each comment word being located at the first position based on the comment word determination model and the plurality of content representation words, the method further includes:

[0019] Determining weight information of the first position based on the probability that each content representation word is located at the first position and the probability that each comment word is located at the first position, wherein the weight information is used to indicate the probability that a content representation word is selected at a corresponding position or the probability that a comment word is selected at a corresponding position;

[0020] The step of determining the selected word at the first position in the comment based on the core word, the probability of each core word being located at the first position, the comment word, and the probability of each comment word being located at the first position comprises:

[0021] Based on the core word, the probability of each core word being located at the first position, the comment word, the probability of each comment word being located at the first position, and the weight information of the first position, a selected word for the first position in the comment is determined.

[0022] In a possible implementation, the comment language analysis module is used to:

[0023] The multiple content representation words, the probability of each content representation word being located at the first position, the comment words and the probability of each comment word being located at the first position are input into a weight information determination model to obtain weight information of the first position.

[0024] In a possible implementation, the comment analysis module is used to:

[0025] Determining, based on the weight information of the first position, a first probability that the content representation word is selected at the first position and a second probability that the comment word is selected at the first position;

[0026] Calculate the product of the probability that each core word is located at the first position and the first probability to obtain a first comprehensive probability corresponding to each core word, and calculate the product of the probability that each comment word is located at the first position and the second probability to obtain a second comprehensive probability corresponding to each comment word;

[0027] Determine the intersection of all core words and all comment words;

[0028] For each core word that does not belong to the intersection, taking the first comprehensive probability corresponding to the core word as the selected probability corresponding to the core word;

[0029] For each comment word that does not belong to the intersection, taking the second comprehensive probability corresponding to the comment word as the selected probability corresponding to the comment word;

[0030] For each word belonging to the intersection, taking the sum of the first comprehensive probability and the second comprehensive probability corresponding to the word as the selected probability corresponding to the word;

[0031] Among all core words and all comment words, the word corresponding to the highest selection probability is determined as the selected word at the first position.

[0032] In a possible implementation, the comment analysis module is used to:

[0033] Determine the category information and multiple content representation words corresponding to the information to be reviewed based on the content analysis model;

[0034] The determining, based on the core word analysis model, the core words in the plurality of content representation words and the probability that each core word is located at the first position in the comment of the information to be commented on comprises:

[0035] Based on the core word analysis model and the category information, determining the core words in the plurality of content representation words and the probability that each core word is located in the first position in the comment of the information to be commented on;

[0036] The step of determining the comment words corresponding to the information to be commented on and the probability of each comment word being located at the first position based on the comment word determination model and the plurality of content representation words includes:

[0037] Based on the review word determination model, the multiple content representation words and the category information, the review words corresponding to the information to be reviewed and the probability of each review word being located at the first position are determined.

[0038] In a possible implementation, the comment analysis module is used to:

[0039] determining a second position in the comment as a target position currently being processed;

[0040] Determine, based on a core word analysis model, the plurality of content representation words, and a selected word at a previous position of the target position, a core word in the plurality of content representation words, and a probability that each core word is located at the target position;

[0041] Determine the comment word corresponding to the information to be commented on and the probability of each comment word being located at the target position based on the comment word determination model, the multiple content representation words and the selected word at the previous position of the target position;

[0042] Determine the selected word at the target position in the comment based on the core word, the probability that each core word is located at the target position, the comment word, and the probability that each comment word is located at the target position;

[0043] If the loop end condition is not met, the position after the target position in the comment is determined as the target position currently being processed, and the process switches to extracting feature information of the selected word before the target position.

[0044] In a possible implementation manner, the information to be commented on includes at least one of text information, image information, and voice information.

[0045] In a second aspect, a device for generating comments is provided, the device comprising:

[0046] A content analysis module, used to determine a plurality of content representation words corresponding to the information to be reviewed based on a content analysis model;

[0047] Comment analysis module, used to:

[0048] Based on the core word analysis model, determining the core words in the plurality of content representation words and the probability that each core word is located in the first position in the comment of the information to be commented on;

[0049] Determine, based on a comment word determination model and the plurality of content representation words, a comment word corresponding to the information to be commented on and a probability that each comment word is located at the first position;

[0050] Determine a selected word at the first position in the comment based on the core word, the probability that each core word is located at the first position, the comment word, and the probability that each comment word is located at the first position;

[0051] Based on the selected word at the first position and the multiple content representation words, the selected words at other positions in the comment are determined, and the selected words at the first position and the selected words at other positions are combined to obtain the comment.

[0052] In a possible implementation, the content analysis module is further used to:

[0053] Based on the probability analysis model, determining the probability that each content representation word is located at the first position;

[0054] The determining, based on the core word analysis model, the core words in the plurality of content representation words and the probability that each core word is located at the first position in the comment of the information to be commented on comprises:

[0055] Based on the core word analysis model, the plurality of content representation words and the probability of each content representation word being located at the first position, a core word in the plurality of content representation words and the probability of each core word being located at the first position are determined.

[0056] In a possible implementation, the comment analysis module is used to:

[0057] Based on the comment word determination model, the plurality of content representation words and the probability of each content representation word being located at the first position, the comment words corresponding to the information to be commented on and the probability of each comment word being located at the first position are determined.

[0058] In a possible implementation, the comment analysis module is further used to:

[0059] Determining weight information of the first position based on the probability that each content representation word is located at the first position and the probability that each comment word is located at the first position, wherein the weight information is used to indicate the probability that a content representation word is selected at a corresponding position or the probability that a comment word is selected at a corresponding position;

[0060] The step of determining the selected word at the first position in the comment based on the core word, the probability of each core word being located at the first position, the comment word, and the probability of each comment word being located at the first position comprises:

[0061] Based on the core word, the probability of each core word being located at the first position, the comment word, the probability of each comment word being located at the first position, and the weight information of the first position, a selected word for the first position in the comment is determined.

[0062] In a possible implementation, the comment analysis module is used to:

[0063] The multiple content representation words, the probability of each content representation word being located at the first position, the comment words and the probability of each comment word being located at the first position are input into a weight information determination model to obtain weight information of the first position.

[0064] In a possible implementation, the comment analysis module is used to:

[0065] Determining, based on the weight information of the first position, a first probability that the content representation word is selected at the first position and a second probability that the comment word is selected at the first position;

[0066] Calculate the product of the probability that each core word is located at the first position and the first probability to obtain a first comprehensive probability corresponding to each core word, and calculate the product of the probability that each comment word is located at the first position and the second probability to obtain a second comprehensive probability corresponding to each comment word;

[0067] Determine the intersection of all core words and all comment words;

[0068] For each core word that does not belong to the intersection, taking the first comprehensive probability corresponding to the core word as the selected probability corresponding to the core word;

[0069] For each comment word that does not belong to the intersection, taking the second comprehensive probability corresponding to the comment word as the selected probability corresponding to the comment word;

[0070] For each word belonging to the intersection, taking the sum of the first comprehensive probability and the second comprehensive probability corresponding to the word as the selected probability corresponding to the word;

[0071] Among all core words and all comment words, the word corresponding to the highest selection probability is determined as the selected word at the first position.

[0072] In a possible implementation, the content analysis module is used to:

[0073] Determine the category information and multiple content representation words corresponding to the information to be reviewed based on the content analysis model;

[0074] The determining, based on the core word analysis model, the core words in the plurality of content representation words and the probability that each core word is located at the first position in the comment of the information to be commented on comprises:

[0075] Based on the core word analysis model and the category information, determining the core words in the plurality of content representation words and the probability that each core word is located in the first position in the comment of the information to be commented on;

[0076] The step of determining the comment word corresponding to the information to be commented on and the probability of each comment word being located at the first position based on the comment word determination model and the plurality of content representation words includes:

[0077] Based on the review word determination model, the multiple content representation words and the category information, the review words corresponding to the information to be reviewed and the probability of each review word being located at the first position are determined.

[0078] In a possible implementation, determining the selected words at other positions in the comment based on the selected words at the first position and the plurality of content representation words includes:

[0079] determining a second position in the comment as a target position currently being processed;

[0080] Determine, based on a core word analysis model, the plurality of content representation words, and a selected word at a previous position of the target position, a core word in the plurality of content representation words, and a probability that each core word is located at the target position;

[0081] Determine the comment word corresponding to the information to be commented on and the probability of each comment word being located at the target position based on the comment word determination model, the multiple content representation words and the selected word at the previous position of the target position;

[0082] Determine the selected word at the target position in the comment based on the core word, the probability that each core word is located at the target position, the comment word, and the probability that each comment word is located at the target position;

[0083] If the loop end condition is not met, the position after the target position in the comment is determined as the target position currently being processed, and the process switches to extracting feature information of the selected word before the target position.

[0084] In a possible implementation manner, the information to be commented on includes at least one of text information, image information, and voice information.

[0085] According to a third aspect, a computer device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, and the instruction is loaded and executed by the processor to implement the operations performed by the method for generating comments.

[0086] In a fourth aspect, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, and the instruction is loaded and executed by a processor to implement the operations performed by the method for generating comments.

[0087] In a fifth aspect, a computer program product is provided, the computer program product comprising a computer program code, and when the computer program code is executed by a computer device, the computer device executes the method of the above-mentioned first aspect and its possible implementation methods.

[0088] The beneficial effects of the technical solution provided by the embodiment of the present application are:

[0089] The scheme mentioned in the embodiment of the present application can determine the probability of the core word and each core word being located at each position of the comment according to the multiple content representation words corresponding to the information to be commented on, and at the same time, determine the probability of the comment word and each comment word being located at each position of the comment, thereby determining the selected word for each position of the comment word and determining the comment corresponding to the information to be commented on. Compared with manual comments, this method improves the comment efficiency and reduces manpower input; compared with template-based automatic comments, words closely related to the information to be commented on can be determined for comment, and there is no need to stick to the requirements of the template sentence pattern, which improves the diversity of the comment sentence pattern and enhances the attractiveness of the comment content to users, thereby facilitating interaction with users. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0091] Figure 1 It is a flow chart of a method for generating comments provided by an embodiment of the present application;

[0092] Figure 2 It is a flow chart of a method for generating comments provided by an embodiment of the present application;

[0093] Figure 3 It is a flow chart of a method for generating comments provided by an embodiment of the present application;

[0094] Figure 4 It is a flow chart of a method for generating comments provided by an embodiment of the present application;

[0095] Figure 5 It is a structural schematic diagram of a device for generating comments provided in an embodiment of the present application;

[0096] Figure 6 It is a structural block diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0097] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0098] The method for generating comments provided in the embodiment of the present application can be used to comment on content posted by users, for example, commenting on social news shared by users in a certain application, commenting on life dynamics posted by users in a certain application, commenting on consumption experiences posted by users in a certain application at a location corresponding to a certain product or a certain merchant, and so on.

[0099] Next, the embodiment of the present application takes the example of a user commenting on a consumption experience posted by a merchant in a certain application to explain in detail the method for generating comments provided by the present application. Other scenarios are similar and will not be described in detail in the embodiment of the present application.

[0100] The embodiment of the present application provides a method for generating comments, and the execution subject of the method may be a server. The server may include a processor, a memory, and a communication component.

[0101] The processor may be a CPU (central processing unit) or a SoC (system on chip), etc. The processor may be used to determine multiple content representation words corresponding to the information to be reviewed, may be used to determine a core word among multiple content representation words, may be used to determine the probability that each core word is located at the first position in the comment language of the information to be reviewed, may be used to determine the comment words corresponding to the information to be reviewed, may be used to determine the probability that each comment word is located at the first position in the comment language, may be used to determine the comment language, and so on.

[0102] The memory may be any volatile memory or non-volatile memory, such as SSD (solid state disk), DRAM (dynamic random access memory), etc. The memory may be used to store pre-stored data, intermediate data, and result data in the process of generating comments. For example, multiple content representation words corresponding to the information to be commented on, core words among the multiple content representation words, the probability that each core word is located in the first position in the comment language of the information to be commented on, the comment words corresponding to the information to be commented on, the probability that each comment word is located in the first position in the comment language, the comment language corresponding to the information to be commented on, and so on.

[0103] The communication component may be a wired network connector, a WiFi (wireless fidelity) module, a Bluetooth module, a cellular network communication module, etc. The communication component may be used to transmit data with other devices, and the other devices may be other servers or operation terminals, etc. For example, the communication component may be used to obtain the information to be commented on sent by the terminal, and may be used to send comments corresponding to the information to be commented on to the terminal, etc.

[0104] Users can select the products they want to consume in the consumer application, and make the corresponding consumption after completing the selection. After completing the consumption, users can enter the post-consumption experience information in the corresponding location of the consumer application. For example, a user purchases a consumption voucher for set meal A1 sold by restaurant A in the consumer application. After the purchase, the user goes to restaurant A and uses the consumption voucher for set meal A1. After the meal, the user enters the post-consumption experience information in a certain location in the application, such as the environment of restaurant A, the service of the service staff, and the taste of the dishes in set meal A.

[0105] After the user enters the above experience information at the designated location of the application, he can click the "Publish" button in the application. The terminal used by the user sends the above experience information to the server. The server receives the above experience information and publishes it in the consumer application, that is, it can be displayed to any user using the consumer application.

[0106] The present application embodiment provides Figure 1 The processing flow of the method for generating comments shown includes the following processing steps:

[0107] S101, obtaining information to be reviewed.

[0108] After acquiring the experience information sent by the user through the terminal, the server determines the experience information as information to be commented on.

[0109] The information to be reviewed may include text information, for example, text entered by the user to express the experience "Kung Pao Chicken is really delicious", "Restaurant A is remote and inconvenient to transport", "The waiters are friendly and enthusiastic", "The environment of Restaurant A is very average", etc.

[0110] The information to be reviewed may also include image information, for example, a picture of a dish taken by a user, a picture of the house number of Restaurant A, a video of the interior environment of Restaurant A, a video of the waiter serving the dish, and so on.

[0111] The information to be reviewed may also include voice information, for example, a recording of a user evaluating the taste of a dish, and so on.

[0112] S102: Determine a plurality of content representation words corresponding to the information to be reviewed based on a content analysis model.

[0113] The server pre-stores a content analysis model. The server inputs the information to be reviewed into the content analysis model to determine a plurality of content representation words corresponding to the information to be reviewed.

[0114] The content analysis model may include a text segmentation module. The server inputs the text information in the information to be reviewed into the text segmentation module, performs segmentation on the text information, and thereby determines the content representation words corresponding to the text information. For example, the server performs segmentation on "Restaurant A opens to reward customers, the big plate chicken is really delicious" through the text segmentation module of the content analysis model, and determines that the corresponding content representation words are "restaurant A", "opening", "rewarding customers", "big plate chicken", "real", "delicious" and "ah", etc.

[0115] The content analysis model may also include an image recognition module. The server inputs the image information in the information to be reviewed into the image recognition model, identifies the entities contained in the image information, and uses the names corresponding to the entities contained in the image information as content representation words. For example, the server uses the text segmentation module of the content analysis model to identify the photo of the dish, and determines that the corresponding dish name is a content representation word, and so on.

[0116] The content analysis model may also include a speech recognition module. The server inputs the speech information in the information to be commented on into the speech recognition module, and may first determine the text information corresponding to the speech information, and then input the text information into the text segmentation module for word segmentation, thereby determining the content representation words corresponding to the speech information.

[0117] Optionally, after the server tokenizes the text information, it can first determine the part of speech corresponding to each word, and then, filter according to the preset part-of-speech conditions, and use the filtered words as the content representation words corresponding to the text information. For example, the preset part-of-speech condition in the server is to use nouns and adjectives in the text information as content representation words. For the text information "Restaurant A is having a grand opening promotion, and the big plate chicken is really delicious", after tokenization, we get "Restaurant A", "opening", "promotion", "big plate chicken", "really", "delicious", and "ah". According to the preset part-of-speech conditions, the content representation words are determined to be "Restaurant A", "big plate chicken", and "delicious", and so on.

[0118] The content analysis model may further include a feature extraction module. After the server determines multiple content representation words through the above one or more modules, it inputs the multiple content representation words into the feature extraction module, thereby determining the feature vector corresponding to each content representation word. Among them, the feature extraction module may be a vector mapping module (also referred to as an Embedding module), etc.

[0119] S103. Based on the probability analysis model, determine the probability that each content representation word is in the first position.

[0120] Among them, the first position refers to the position of the first word of the comment corresponding to the information to be commented. A comment consists of multiple words, and each word corresponds to a position. For example, if the comment is "The big plate chicken looks really fragrant", then "big plate chicken" is in the first position of the comment, "looks" is in the second position of the comment, "really" is in the third position of the comment, and "fragrant" is in the fourth position of the comment.

[0121] The probability analysis model is pre-stored in the server. Among them, the probability analysis model may be a self-attention model, such as a model with a Transformer structure, etc. The server inputs the feature vector corresponding to each content representation word into the probability analysis model to obtain the probability that each content representation word is in the first position, that is, the probability that each content representation word is the first word of the comment.

[0122] S104. Based on the core word analysis model and the probability that each content representation word is in the first position, determine the core words among the multiple content representation words, and the probability that each core word is in the first position.

[0123] The server pre-stores a core word analysis model. The server inputs the feature vector corresponding to each content representation word and the probability of each content representation word being in the first position into the core word analysis model, and can determine the core words among multiple content representation words and the probability of each core word being in the first position. For example, the multiple content representation words are "big plate chicken", "look", "起来 (up)", "really fragrant", and the probabilities of each content representation word being in the first position are 50%, 30%, 15%, and 5% respectively. Input the feature vectors corresponding to the multiple content representation words and the probability of each content representation word being in the first position into the core word analysis model, and determine that the core words are "big plate chicken" and "really fragrant", and the probabilities of each core word being in the first position are 50% and 5% respectively.

[0124] Optionally, the core word analysis model can recalculate the probability of each core word being in the first position. For example, the content representation words are "big plate chicken", "look", "起来 (up)" and "really fragrant", and the probabilities of each content representation word being in the first position are 50%, 30%, 15%, and 5% respectively. The server determines through the core word analysis model that the core words are "big plate chicken" and "really fragrant", and the probabilities of each core word being in the first position are 91% and 9% respectively.

[0125] S105. Based on the comment word determination model, multiple content representation words, and the probability of each content representation word being in the first position, determine the comment words corresponding to the to-be-commented information and the probability of each comment word being in the first position.

[0126] The server pre-stores a comment word determination model and a comment word library. Among them, the comment word determination model can be an LSTM-RNN (Long Short Term Memory-Recurrent Neural Network) model, or a Transformer-Decoder (Transformer-Decoder) model, etc. The server inputs the feature vector corresponding to each content representation word and the probability of each content representation word being in the first position into the comment word determination model, and determines the probability of each comment word in the comment word library being in the first position.

[0127] Optionally, the server may first determine the probability of each comment word in the comment word library being located in the first position through the comment word determination model; then, perform screening according to the pre-set comment word screening conditions; finally, determine the comment words that meet the conditions as the comment words corresponding to the information to be reviewed, and determine the probability of each comment word that meets the conditions being located in the first position. The comment word screening conditions may be all comment words that meet a certain probability threshold, for example, determine all comment words with a probability greater than 10% as the comment words corresponding to the information to be reviewed, and so on. The comment word screening conditions may also be all comment words that meet a certain quantity threshold, for example, determine the 10 comment words with the highest probability as the comment words corresponding to the information to be reviewed, and so on. The comment word screening conditions may also be set according to the part of speech of the comment words, for example, for nouns, determine all comment words with a probability greater than 20% as the comment words corresponding to the information to be reviewed, for verbs, determine the 5 comment words with the highest probability as the comment words corresponding to the information to be reviewed, and so on. The comment word screening conditions are only exemplarily described here, without limiting them, and other screening conditions are not described here.

[0128] S106 , determining a selected word at the first position in the comment based on the core word, the probability of each core word being at the first position, the comment word, and the probability of each comment word being at the first position.

[0129] First, the server determines the intersection of all the above core words and all the above comment words.

[0130] Then, the server determines the selected probability of each word being located at the first position based on the probability of each comment word being located at the first position and the probability of each core word being located at the first position.

[0131] For each core word that does not belong to the above intersection, the probability of the core word being in the first position is taken as the selected probability corresponding to the core word. For each comment word that does not belong to the above intersection, the probability of the comment word being in the first position is taken as the selected probability corresponding to the comment word. For each word that belongs to the above intersection, the sum of the probability of the word being in the first position when used as a core word and the probability of the word being in the first position when used as a comment word is determined as the selected probability corresponding to the word.

[0132] Finally, the server determines the selected word for the first position of the review according to the selected probability corresponding to each word. The server may determine the word corresponding to the highest selected probability as the selected word for the first position.

[0133] Optionally, the server may pre-store probability coefficients corresponding to different parts of speech at the first position. For words of different parts of speech, the server may use the product of the selection probability of the word and the probability coefficient as the total selection probability corresponding to the word, and then the server determines the word corresponding to the highest total selection probability as the selected word at the first position. This is only an exemplary description of the determination process of the selected word, without limiting it, and other determination methods are not described in detail here.

[0134] After the server determines the comment word at the first position of the comment, it makes a judgment based on the pre-set comment end condition, where the comment end condition can be a length threshold of the field contained in the comment, or can be a part of speech end condition, etc. If the end condition is met, the comment word at the first position is determined as the comment corresponding to the information to be reviewed. If the end condition is not met, the selected word at the second position after the first position is determined. Next, the process of determining the selected word at the second position of the comment is explained.

[0135] S107: Determine the second position in the comment as the target position for current processing.

[0136] First, the server inputs the feature information corresponding to each content representation word and the feature information of the selected word at the previous position of the target position (i.e., the selected word at the first position) into the probability analysis model to determine the probability of each content representation word being located at the target position (i.e., the second position) being processed currently. The specific process is similar to step S103 and will not be described in detail here.

[0137] Then, the server inputs the feature information corresponding to each content representation word and the probability of each content representation word being located at the second position into the core word determination model to determine the core word corresponding to the second position of the multiple content representation words and the probability of each core word being located at the second position. The specific process is similar to step S104 and will not be described in detail here.

[0138] At the same time, the server inputs the feature information corresponding to each content representation word, the probability of each content representation word being located at the second position, and the feature information of the selected word at the previous position of the target position (i.e., the selected word at the first position) into the comment word determination model to determine the comment word corresponding to the second position of the information to be commented on, and the probability of each comment word being located at the second position. The specific process is similar to step S105 and will not be repeated here.

[0139] Finally, the server determines the selected word at the second position in the comment based on the core word, the probability of each core word being at the target position, the comment word, and the probability of each comment word being at the target position. The specific process is similar to step S106 and will not be described here.

[0140] After the server determines the selected word at the second position, it makes a judgment according to the pre-set comment end condition. If the comment end condition is met, the selected word at the first position and the selected word at the second position are combined to obtain the comment corresponding to the information to be commented on. If the loop end condition is not met, the position after the target position in the comment is determined as the target position currently being processed, and step S107 is continued until the loop end condition is met.

[0141] When the loop end condition is met, the server combines the selected words in all positions to obtain the comment corresponding to the information to be commented. The loop process described in the above steps S102 to S107 is as follows: Figure 2 shown.

[0142] S108: Send the comments corresponding to the information to be commented on to the terminal.

[0143] After the server determines the comment corresponding to the information to be reviewed, it will publish the comment at the location corresponding to the information to be reviewed in the application, and send a comment reminder to the user through the terminal to remind the user to check the comment. At the same time, the information to be reviewed and the corresponding comment can be displayed to any user using the application.

[0144] The present application embodiment provides Figure 3 The processing flow of the method for generating comments shown in FIG. 1 can calculate the weight corresponding to each delivery stage, including the following processing steps:

[0145] S301, obtaining information to be reviewed.

[0146] The server obtains at least one of the text information, image information and voice information input by the user as the information to be commented on. The specific description is the same as step S101 and will not be repeated here.

[0147] S302: Determine multiple content representation words corresponding to the information to be reviewed based on the content analysis model.

[0148] The server pre-stores a content analysis model, and inputs the information to be reviewed into the content analysis model to determine multiple content representation words corresponding to the information to be reviewed and feature information (i.e., feature vector) corresponding to each content representation word. The specific process is the same as step S102, and will not be repeated here.

[0149] S303: Determine the probability of each content representation word being located at the first position based on the probability analysis model.

[0150] The server inputs the feature vector corresponding to each content representation word into the probability analysis model to obtain the probability of each content representation word being located at the first position. The specific process is the same as step S103 and will not be described in detail here.

[0151] S304 : Determine core words in the plurality of content representation words and the probability of each core word being located at the first position based on the core word analysis model and the probability of each content representation word being located at the first position.

[0152] The server pre-stores a core word analysis model. The server inputs the feature vector corresponding to each content representation word and the probability of each content representation word being located at the first position into the core word analysis model, and can determine the core word among multiple content representation words and the probability of each core word being located at the first position. The specific process is the same as step S104 and will not be repeated here.

[0153] S305 : Determine the comment word corresponding to the information to be commented on and the probability of each comment word being located at the first position based on the comment word determination model, the multiple content representation words and the probability of each content representation word being located at the first position.

[0154] The server pre-stores a review word determination model and a review word library. The server inputs the feature vector corresponding to each content representation word and the probability of each content representation word being located at the first position into the review word determination model to determine the probability of each review word in the review word library being located at the first position. The specific process is the same as step S105 and will not be repeated here.

[0155] S306: Determine weight information of the first position based on the probability that each content representation word is located at the first position and the probability that each comment word is located at the first position.

[0156] The server pre-stores a weight information determination model, which may be a binary classification model. The weight information model is used to determine the weight information of the target position, and the weight information is used to indicate the probability of the content representation word being selected at the target position and / or the probability of the comment word being selected at the target position, wherein the sum of the probability of the content representation word being selected at the target position and the probability of the comment word being selected at the target position is equal to 1.

[0157] The server inputs the feature information corresponding to each content characterization word, the probability of each content characterization word being in the first position, the feature information corresponding to each comment word, and the probability of each comment word being in the first position into the weight information determination model to obtain the weight information of the first position.

[0158] S307 : Based on the weight information of the first position, determine a first probability that the content representation word is selected at the first position and a second probability that the comment word is selected at the first position.

[0159] First, the server determines a first probability that the content representation word is selected at the first position and a second probability that the comment word is selected at the first position according to the weight information of the first position.

[0160] Then, the server calculates the product of the probability of each core word being in the first position and the first probability, and obtains the first comprehensive probability corresponding to each core word. At the same time, the server calculates the product of the probability of each comment word being in the first position and the second probability, and obtains the second comprehensive probability corresponding to each comment word.

[0161] Finally, the server determines the selection probability of each word being located at the first position according to the first comprehensive probability corresponding to each core word and the second comprehensive probability corresponding to each comment word. The selection probability determination process is as follows:

[0162] The server determines the intersection of all core words and all comment words. For each core word that does not belong to the above intersection, the first comprehensive probability corresponding to the core word is used as the selected probability corresponding to the core word. For each comment word that does not belong to the above intersection, the second comprehensive probability corresponding to the comment word is used as the selected probability corresponding to the comment word. For each word that belongs to the above intersection, the sum of the first comprehensive probability and the second comprehensive probability corresponding to the word is used as the selected probability corresponding to the word. Furthermore, among all core words and all comment words, the word with the highest selected probability is determined as the selected word in the first position. The specific process of determining the selected probability is the same as step S106 and will not be repeated here.

[0163] After the server determines the comment word at the first position of the comment, it makes a judgment according to the pre-set comment ending condition. If the ending condition is met, the comment word at the first position is determined as the comment corresponding to the information to be reviewed. If the ending condition is not met, the server continues to determine the selected word at the second position after the first position. Next, the process of determining the selected word at the second position of the comment is described.

[0164] S308: Determine the second position in the comment as the target position for current processing.

[0165] First, the server inputs the feature information corresponding to each content representation word and the feature information of the selected word at the previous position of the target position (i.e., the selected word at the first position) into the probability analysis model to determine the probability of each content representation word being located at the target position (i.e., the second position) being processed currently. The specific process is similar to step S103 and will not be described in detail here.

[0166] Second, the server inputs the feature information corresponding to each content representation word and the probability of each content representation word being located at the second position into the core word determination model to determine the core word corresponding to the second position of the multiple content representation words and the probability of each core word being located at the second position. The specific process is similar to step S104 and will not be described in detail here.

[0167] Third, the server inputs the feature information corresponding to each content representation word, the probability of each content representation word being located at the second position, and the feature information of the selected word at the previous position of the target position (i.e., the selected word at the first position) into the comment word determination model to determine the comment word corresponding to the second position of the information to be commented on, and the probability of each comment word being located at the second position. The specific process is similar to step S105 and will not be described in detail here.

[0168] Fourth, the server inputs the feature information corresponding to each content characterization word, the probability of each content characterization word being located at the second position, the feature information corresponding to each comment word, and the probability of each comment word being located at the second position into the weight information determination model to obtain the weight information of the second position. The specific process is similar to step S306 and will not be repeated here.

[0169] Fifth, the server first determines the third probability of the content representation word being selected in the second position and the fourth probability of the comment word being selected in the second position based on the weight information of the first position. Then, the server calculates the third comprehensive probability corresponding to each core word and the fourth comprehensive probability corresponding to each comment word. Finally, the server determines the selection probability of each word in the second position based on the third comprehensive probability corresponding to each core word and the fourth comprehensive probability corresponding to each comment word, thereby determining the selected word in the second position. The specific process is similar to step S307 and will not be repeated here.

[0170] After the server determines the selected word at the second position, it makes a judgment according to the pre-set comment end condition. If the comment end condition is met, the selected word at the first position and the selected word at the second position are combined to obtain the comment corresponding to the information to be commented on. If the loop end condition is not met, the position after the target position in the comment is determined as the target position currently being processed, and step S208 is continued until the loop end condition is met.

[0171] When the loop end condition is met, the server combines the selected words in all positions to obtain the comment corresponding to the information to be commented. Figure 4 shown.

[0172] S309: Send the comments corresponding to the information to be commented on to the terminal.

[0173] After the server determines the comment corresponding to the information to be reviewed, it will publish the comment at the location corresponding to the information to be reviewed in the application, and send a comment reminder to the user through the terminal to remind the user to check the comment. At the same time, the information to be reviewed and the corresponding comment can be displayed to any user using the application.

[0174] Optionally, the server inputs the information to be reviewed into the content analysis model, and can determine the category information and multiple content representation words corresponding to the information to be reviewed. The category information is used to indicate the category to which the information to be reviewed belongs, for example, bathing, hot pot, cooking, indoor entertainment, outdoor leisure, life skills, knowledge sharing, etc. The above category information can be set by the application developer, and will not be elaborated here.

[0175] When the server executes the above steps S103, S104, S105, S107, S303, S304, S305, S306, and S308, it is necessary to input the feature vector corresponding to the above category information and the feature vector corresponding to each content representation word into the corresponding model for corresponding processing. For example, the server inputs the feature vector corresponding to the category information, the feature vector corresponding to each content representation word, and the probability that each content representation word is located in the first position into the core word analysis model to determine the core word among multiple content representation words and the probability that each core word is located in the first position. For another example, the server inputs the feature vector corresponding to the category information, the feature vector corresponding to each content representation word, and the probability that each content representation word is located in the first position into the comment word determination model to determine the comment word corresponding to the information to be commented on and the probability that each comment word is located in the first position. Other steps that use the feature vector corresponding to the category information as input will not be repeated here.

[0176] All the above optional technical solutions can be arbitrarily combined to form optional embodiments of the present application, which will not be described one by one here.

[0177] The scheme mentioned in the embodiment of the present application can first determine multiple content representation words corresponding to the information to be commented on; then, based on the multiple content representation words, determine the core words and the probability of each core word being located in the first position of the comment, and at the same time, determine the comment words and the probability of each comment word being located in the first position of the comment; then, determine the selected words in the first position of the comment, and based on the selected words in the first position and the content representation words, the selected words in other positions of the comment; finally, based on the selected words in each position, determine the comment corresponding to the information to be commented on. Compared with manual comments, this method improves the comment efficiency and reduces manpower input; compared with template-based automatic comments, it can determine words closely related to the information to be commented on for commenting, and does not need to stick to the requirements of the template sentence pattern, which improves the diversity of the comment sentence pattern and enhances the attractiveness of the comment content to users, thereby facilitating interaction with users.

[0178] The present application embodiment provides a device for determining an estimated delivery time, which may be the server in the above embodiment, such as Figure 5 As shown, the device comprises:

[0179] A content analysis module 510, configured to determine a plurality of content representation words corresponding to the information to be reviewed based on a content analysis model;

[0180] The comment analysis module 520 is used to:

[0181] Based on the core word analysis model, determining the core words in the plurality of content representation words and the probability that each core word is located in the first position in the comment of the information to be commented on;

[0182] Determine, based on a comment word determination model and the plurality of content representation words, a comment word corresponding to the information to be commented on and a probability that each comment word is located at the first position;

[0183] Determine a selected word at the first position in the comment based on the core word, the probability that each core word is located at the first position, the comment word, and the probability that each comment word is located at the first position;

[0184] Based on the selected word at the first position and the multiple content representation words, the selected words at other positions in the comment are determined, and the selected words at the first position and the selected words at other positions are combined to obtain the comment.

[0185] In a possible implementation, the content analysis module 510 is further configured to:

[0186] Based on the probability analysis model, determining the probability that each content representation word is located at the first position;

[0187] The determining, based on the core word analysis model, the core words in the plurality of content representation words and the probability that each core word is located at the first position in the comment of the information to be commented on comprises:

[0188] Based on the core word analysis model, the plurality of content representation words and the probability of each content representation word being located at the first position, a core word in the plurality of content representation words and the probability of each core word being located at the first position are determined.

[0189] In a possible implementation, the comment analysis module 520 is used to:

[0190] Based on the comment word determination model, the plurality of content representation words and the probability of each content representation word being located at the first position, the comment words corresponding to the information to be commented on and the probability of each comment word being located at the first position are determined.

[0191] In a possible implementation, the comment analysis module 520 is further used to:

[0192] Determining weight information of the first position based on the probability that each content representation word is located at the first position and the probability that each comment word is located at the first position, wherein the weight information is used to indicate the probability that a content representation word is selected at a corresponding position or the probability that a comment word is selected at a corresponding position;

[0193] The step of determining the selected word at the first position in the comment based on the core word, the probability of each core word being located at the first position, the comment word, and the probability of each comment word being located at the first position comprises:

[0194] Based on the core word, the probability of each core word being located at the first position, the comment word, the probability of each comment word being located at the first position, and the weight information of the first position, a selected word for the first position in the comment is determined.

[0195] In a possible implementation, the comment analysis module 520 is used to:

[0196] The multiple content representation words, the probability of each content representation word being located at the first position, the comment words and the probability of each comment word being located at the first position are input into a weight information determination model to obtain weight information of the first position.

[0197] In a possible implementation, the comment analysis module 520 is used to:

[0198] Determining, based on the weight information of the first position, a first probability that the content representation word is selected at the first position and a second probability that the comment word is selected at the first position;

[0199] Calculate the product of the probability that each core word is located at the first position and the first probability to obtain a first comprehensive probability corresponding to each core word, and calculate the product of the probability that each comment word is located at the first position and the second probability to obtain a second comprehensive probability corresponding to each comment word;

[0200] Determine the intersection of all core words and all comment words;

[0201] For each core word that does not belong to the intersection, taking the first comprehensive probability corresponding to the core word as the selected probability corresponding to the core word;

[0202] For each comment word that does not belong to the intersection, taking the second comprehensive probability corresponding to the comment word as the selected probability corresponding to the comment word;

[0203] For each word belonging to the intersection, taking the sum of the first comprehensive probability and the second comprehensive probability corresponding to the word as the selected probability corresponding to the word;

[0204] Among all core words and all comment words, the word corresponding to the highest selection probability is determined as the selected word at the first position.

[0205] In a possible implementation, the content analysis module 510 is used to:

[0206] Determine the category information and multiple content representation words corresponding to the information to be reviewed based on the content analysis model;

[0207] The determining, based on the core word analysis model, the core words in the plurality of content representation words and the probability that each core word is located at the first position in the comment of the information to be commented on comprises:

[0208] Based on the core word analysis model and the category information, determining the core words in the plurality of content representation words and the probability that each core word is located in the first position in the comment of the information to be commented on;

[0209] The step of determining the comment words corresponding to the information to be commented on and the probability of each comment word being located at the first position based on the comment word determination model and the plurality of content representation words includes:

[0210] Based on the review word determination model, the multiple content representation words and the category information, the review words corresponding to the information to be reviewed and the probability of each review word being located at the first position are determined.

[0211] In a possible implementation, the comment analysis module 520 is used to:

[0212] determining a second position in the comment as a target position currently being processed;

[0213] Determine, based on a core word analysis model, the plurality of content representation words, and a selected word at a previous position of the target position, a core word in the plurality of content representation words, and a probability that each core word is located at the target position;

[0214] Determine the comment word corresponding to the information to be commented on and the probability of each comment word being located at the target position based on the comment word determination model, the multiple content representation words and the selected word at the previous position of the target position;

[0215] Determine the selected word at the target position in the comment based on the core word, the probability that each core word is located at the target position, the comment word, and the probability that each comment word is located at the target position;

[0216] If the loop end condition is not met, the position after the target position in the comment is determined as the target position currently being processed, and the process switches to extracting feature information of the selected word before the target position.

[0217] In a possible implementation manner, the information to be commented on includes at least one of text information, image information, and voice information.

[0218] The scheme mentioned in the embodiment of the present application can determine the probability of the core word and each core word being located at each position of the comment according to the multiple content representation words corresponding to the information to be commented on, and at the same time, determine the probability of the comment word and each comment word being located at each position of the comment, thereby determining the selected word for each position of the comment word and determining the comment corresponding to the information to be commented on. Compared with manual comments, this method improves the comment efficiency and reduces manpower input; compared with template-based automatic comments, words closely related to the information to be commented on can be determined for comment, and there is no need to stick to the requirements of the template sentence pattern, which improves the diversity of the comment sentence pattern and enhances the attractiveness of the comment content to users, thereby facilitating interaction with users.

[0219] It should be noted that: the device for generating comments provided in the above embodiment only uses the division of the above functional modules as an example when determining the comment corresponding to the information to be commented on. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the device for generating comments provided in the above embodiment and the method embodiment for generating comments belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0220] An embodiment of the present application provides a computer device, which may be the server in the above embodiment. Figure 6 6 is a schematic diagram of the structure of the computer device. The computer device 600 may have relatively large differences due to different configurations or performances, and may include one or more CPUs (central processing units, processors) 610 and one or more memories 620, wherein the memory 620 stores at least one instruction, and the at least one instruction is loaded and executed by the processor 610 to implement the methods provided by the above-mentioned various method embodiments. Of course, the computer device may also have components such as a wired or wireless network interface, a keyboard, and an input and output interface for input and output. The computer device may also include other components for implementing device functions, which will not be described in detail here.

[0221] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including instructions, which can be executed by a processor in a terminal to complete the method of generating comments in the above embodiment. The computer-readable storage medium can be non-transitory. For example, the computer-readable storage medium can be a ROM (read-only memory), a RAM (random access memory), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0222] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0223] It should be noted that all actions of acquiring signals, information or data in this application are carried out in compliance with the relevant data protection laws and policies of the country where they are located and with the authorization given by the owner of the corresponding device.

[0224] The above description is only an optional embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for generating comments, characterized in that: The method comprises: Determine multiple content representation words corresponding to the information to be reviewed based on the content analysis model; Based on the core word analysis model, determining the core words in the plurality of content representation words and the probability that each core word is located in the first position in the comment of the information to be commented on; Determine, based on a comment word determination model and the plurality of content representation words, a comment word corresponding to the information to be commented on and a probability that each comment word is located at the first position; Determine a selected word at the first position in the comment based on the core word, the probability that each core word is located at the first position, the comment word, and the probability that each comment word is located at the first position; Based on the selected word at the first position and the plurality of content representation words, determine the selected words at other positions in the comment, and combine the selected word at the first position with the selected words at other positions to obtain the comment; After determining the plurality of content representation words corresponding to the information to be reviewed based on the content analysis model, the method further includes: Based on the probability analysis model, determining the probability that each content representation word is located at the first position; After determining the comment words corresponding to the information to be commented on and the probability of each comment word being located at the first position based on the comment word determination model and the plurality of content representation words, the method further includes: Determining weight information of the first position based on the probability that each content representation word is located at the first position and the probability that each comment word is located at the first position, wherein the weight information is used to indicate the probability that a content representation word is selected at a corresponding position or the probability that a comment word is selected at a corresponding position; The step of determining the selected word at the first position in the comment based on the core word, the probability of each core word being located at the first position, the comment word, and the probability of each comment word being located at the first position comprises: Determine the selected word at the first position in the comment based on the core word, the probability of each core word being located at the first position, the comment word, the probability of each comment word being located at the first position, and the weight information of the first position; The step of determining the selected word at the first position in the comment based on the core word, the probability of each core word being located at the first position, the comment word, the probability of each comment word being located at the first position, and the weight information of the first position comprises: Determining, based on the weight information of the first position, a first probability that the content representation word is selected at the first position and a second probability that the comment word is selected at the first position; Calculate the product of the probability that each core word is located at the first position and the first probability to obtain a first comprehensive probability corresponding to each core word, and calculate the product of the probability that each comment word is located at the first position and the second probability to obtain a second comprehensive probability corresponding to each comment word; Determine the intersection of all core words and all comment words; For each core word that does not belong to the intersection, taking the first comprehensive probability corresponding to the core word as the selected probability corresponding to the core word; For each comment word that does not belong to the intersection, taking the second comprehensive probability corresponding to the comment word as the selected probability corresponding to the comment word; For each word belonging to the intersection, taking the sum of the first comprehensive probability and the second comprehensive probability corresponding to the word as the selected probability corresponding to the word; Among all core words and all comment words, the word corresponding to the highest selection probability is determined as the selected word at the first position; The step of determining selected words at other positions in the comment based on the selected words at the first position and the plurality of content representation words includes: determining a second position in the comment as a target position currently being processed; Determine, based on a core word analysis model, the plurality of content representation words, and a selected word at a previous position of the target position, a core word in the plurality of content representation words, and a probability that each core word is located at the target position; Determine the comment word corresponding to the information to be commented on and the probability of each comment word being located at the target position based on the comment word determination model, the multiple content representation words and the selected word at the previous position of the target position; Determine the selected word at the target position in the comment based on the core word, the probability that each core word is located at the target position, the comment word, and the probability that each comment word is located at the target position; If the loop end condition is not met, the position after the target position in the comment is determined as the target position currently being processed, and the process switches to extracting feature information of the selected word before the target position.

2. The method according to claim 1, characterized in that The determining, based on the core word analysis model, the core words in the plurality of content representation words and the probability that each core word is located at the first position in the comment of the information to be commented on comprises: Based on the core word analysis model, the plurality of content representation words and the probability of each content representation word being located at the first position, a core word in the plurality of content representation words and the probability of each core word being located at the first position are determined.

3. The method according to claim 2, characterized in that: The step of determining the comment words corresponding to the information to be commented on and the probability of each comment word being located at the first position based on the comment word determination model and the plurality of content representation words includes: Based on the comment word determination model, the plurality of content representation words and the probability of each content representation word being located at the first position, the comment words corresponding to the information to be commented on and the probability of each comment word being located at the first position are determined.

4. The method according to claim 1, characterized in that: The determining the weight information of the first position based on the probability that each content representation word is located at the first position and the probability that each comment word is located at the first position includes: The multiple content representation words, the probability of each content representation word being located at the first position, the comment words and the probability of each comment word being located at the first position are input into a weight information determination model to obtain weight information of the first position.

5. The method according to claim 1, characterized in that: The determining of a plurality of content representation words corresponding to the information to be reviewed based on the content analysis model includes: Determine the category information and multiple content representation words corresponding to the information to be reviewed based on the content analysis model; The determining, based on the core word analysis model, the core words in the plurality of content representation words and the probability that each core word is located at the first position in the comment of the information to be commented on comprises: Based on the core word analysis model and the category information, determining the core words in the plurality of content representation words and the probability that each core word is located in the first position in the comment of the information to be commented on; The step of determining the comment words corresponding to the information to be commented on and the probability of each comment word being located at the first position based on the comment word determination model and the plurality of content representation words includes: Based on the review word determination model, the multiple content representation words and the category information, the review words corresponding to the information to be reviewed and the probability of each review word being located at the first position are determined.

6. The method according to claim 1, characterized in that The information to be commented on includes at least one of text information, image information, and voice information.

7. A device for generating comments, characterized in that: The device comprises: A content representation module, used to determine a plurality of content representation words corresponding to the information to be reviewed based on a content analysis model; Comment analysis module, used to: Based on the core word analysis model, determining the core words in the plurality of content representation words and the probability that each core word is located in the first position in the comment of the information to be commented on; Determine, based on a comment word determination model and the plurality of content representation words, a comment word corresponding to the information to be commented on and a probability that each comment word is located at the first position; Determine a selected word at the first position in the comment based on the core word, the probability that each core word is located at the first position, the comment word, and the probability that each comment word is located at the first position; Based on the selected word at the first position and the plurality of content representation words, determine the selected words at other positions in the comment, and combine the selected word at the first position with the selected words at other positions to obtain the comment; After determining the plurality of content representation words corresponding to the information to be reviewed based on the content analysis model, the method further includes: Based on the probability analysis model, determining the probability that each content representation word is located at the first position; After determining the comment words corresponding to the information to be commented on and the probability of each comment word being located at the first position based on the comment word determination model and the plurality of content representation words, the method further includes: Determining weight information of the first position based on the probability that each content representation word is located at the first position and the probability that each comment word is located at the first position, wherein the weight information is used to indicate the probability that a content representation word is selected at a corresponding position or the probability that a comment word is selected at a corresponding position; The step of determining the selected word at the first position in the comment based on the core word, the probability of each core word being located at the first position, the comment word, and the probability of each comment word being located at the first position comprises: Determine the selected word at the first position in the comment based on the core word, the probability of each core word being located at the first position, the comment word, the probability of each comment word being located at the first position, and the weight information of the first position; The step of determining the selected word at the first position in the comment based on the core word, the probability of each core word being located at the first position, the comment word, the probability of each comment word being located at the first position, and the weight information of the first position comprises: Determining, based on the weight information of the first position, a first probability that the content representation word is selected at the first position and a second probability that the comment word is selected at the first position; Calculate the product of the probability that each core word is located at the first position and the first probability to obtain a first comprehensive probability corresponding to each core word, and calculate the product of the probability that each comment word is located at the first position and the second probability to obtain a second comprehensive probability corresponding to each comment word; Determine the intersection of all core words and all comment words; For each core word that does not belong to the intersection, taking the first comprehensive probability corresponding to the core word as the selected probability corresponding to the core word; For each comment word that does not belong to the intersection, taking the second comprehensive probability corresponding to the comment word as the selected probability corresponding to the comment word; For each word belonging to the intersection, taking the sum of the first comprehensive probability and the second comprehensive probability corresponding to the word as the selected probability corresponding to the word; Among all core words and all comment words, the word corresponding to the highest selection probability is determined as the selected word at the first position; The step of determining selected words at other positions in the comment based on the selected words at the first position and the plurality of content representation words includes: determining a second position in the comment as a target position currently being processed; Determine, based on a core word analysis model, the plurality of content representation words, and a selected word at a previous position of the target position, a core word in the plurality of content representation words, and a probability that each core word is located at the target position; Determine the comment word corresponding to the information to be commented on and the probability of each comment word being located at the target position based on the comment word determination model, the multiple content representation words and the selected word at the previous position of the target position; Determine the selected word at the target position in the comment based on the core word, the probability that each core word is located at the target position, the comment word, and the probability that each comment word is located at the target position; If the loop end condition is not met, the position after the target position in the comment is determined as the target position currently being processed, and the process switches to extracting feature information of the selected word before the target position.

8. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the operation performed by the method for generating comments according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The storage medium stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the operation performed by the method for generating comments as described in any one of claims 1 to 6.

Citation Information

Patent Citations

  • Chapter content tiering method and device, and article content tiering method and device

    CN103034656A

  • Method and device for analyzing user demands

    CN107102993A

  • Comment viewpoint theme extraction method, electronic equipment and storage medium

    CN112131863A

  • Comment information generation method and device, electronic equipment and storage medium

    CN112667780A

  • Techniques for identifying ingenuine online reviews

    US20200104887A1