Comment display method, device and equipment based on large model, medium and intelligent agent

Abstract

The invention provides a comment display method, device and equipment based on a large model, a medium and an intelligent agent, and relates to the technical field of artificial intelligence, in particular to the fields of large models, generative models, model fine tuning, intelligent recommendation and the like. The specific implementation scheme is as follows: in response to determining that a target resource is a hotspot resource, determining a target style matched with the target resource; based on the large model, according to the attribute information of the target resource and the target style, generating a target comment which aims at the target resource and accords with the target style; and displaying the target comment.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to the fields of large models, generative models, model fine-tuning, intelligent recommendation, etc. More specifically, the present disclosure provides a comment display method, device, electronic device, storage medium, computer program product and intelligent agent based on a large model. Background Art

[0002] With the rapid advancement of Internet technology and the widespread popularity of mobile devices, users can browse resources through the Internet and express their personal opinions on the resources. Summary of the invention

[0003] The present disclosure provides a review display method, device, electronic device, storage medium, computer program product and intelligent agent based on a large model.

[0004] According to one aspect of the present disclosure, a comment display method based on a big model is provided, including: in response to determining that a target resource is a hot resource, determining a target style that matches the target resource; based on the big model, generating a target comment for the target resource and conforming to the target style according to the attribute information and the target style of the target resource; and displaying the target comment.

[0005] According to another aspect of the present disclosure, a comment display device based on a large model is provided, comprising: a style determination module, a first generation module and a first display module. The style determination module is used to determine a target style that matches the target resource in response to determining that the target resource is a hot resource. The first generation module is used to generate a target comment for the target resource and conforming to the target style based on the large model and according to the attribute information and target style of the target resource. The first display module is used to display the target comment.

[0006] According to another aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method provided by the present disclosure.

[0007] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable a computer to execute the method provided by the present disclosure.

[0008] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the method provided by the present disclosure is implemented.

[0009] According to another aspect of the present disclosure, an intelligent agent is provided, comprising: an input module, a processing module and an output module. The input module is used to receive input information. The processing module is used to obtain output information by calling the large model to execute the above method based on the input information received by the input module. The output module is used to output the output information obtained by the processing module.

[0010] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0012] Figure 1 is a schematic diagram of an application scenario of the comment display method and device according to an embodiment of the present disclosure;

[0013] Figure 2 is a schematic flow chart of a comment display method according to an embodiment of the present disclosure;

[0014] Figure 3 is a schematic diagram of the training process of a large model according to an embodiment of the present disclosure;

[0015] Figure 4 is a schematic diagram of a comment display method according to an embodiment of the present disclosure;

[0016] Figure 5 is a schematic structural block diagram of a comment display device according to an embodiment of the present disclosure;

[0017] Figure 6 A schematic diagram of a structure of an intelligent agent according to an embodiment of the present disclosure is shown; and

[0018] Figure 7 It is a structural block diagram of an electronic device used to implement the comment display method of the embodiment of the present disclosure. DETAILED DESCRIPTION

[0019] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0020] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0021] In the technical solution of the present disclosure, the user's authorization or consent is obtained before obtaining or collecting the user's personal information.

[0022] With the rapid advancement of Internet technology and the widespread popularity of mobile devices, users can browse resources through the Internet and express their personal opinions on resources. Some resource platforms will guide users to participate in discussions through pre-configured copywriting, but this method lacks personalization and pertinence, and the user experience is poor. Some operators of resource platforms will comment on resources to enrich the content of the comment area and increase user interaction interest. However, this comment method requires manual writing and replying, which is not only time-consuming and labor-intensive, but also difficult to cope with a large number of comment demands.

[0023] The disclosed embodiments aim to provide a comment display method, which can generate personalized comments, avoid the monotony of comment style and content, improve user experience, reduce labor costs, and improve the efficiency of generating comments.

[0024] The technical solution provided by the present disclosure will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] Figure 1 It is a schematic diagram of an application scenario of the comment display method and device according to an embodiment of the present disclosure.

[0026] It should be noted that Figure 1 What is shown is merely an example of a system architecture to which the embodiments of the present disclosure can be applied, in order to help those skilled in the art understand the technical content of the present disclosure, but it does not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios.

[0027] like Figure 1 As shown, the system architecture 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0028] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Terminal devices 101, 102, 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptops, desktop computers, etc.

[0029] The server 105 may be a server that provides various services, such as a background management server (only an example) that provides support for websites browsed by users using the terminal devices 101, 102, and 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as response data generated based on historical comments input by users, or target comments generated based on attribute information of target resources) to the terminal device.

[0030] It should be noted that the comment display method provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the comment display device provided in the embodiment of the present disclosure can generally be set in the server 105. The comment display method provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105. Accordingly, the comment display device provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105.

[0031] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0032] Figure 2 It is a schematic flowchart of the comment display method according to an embodiment of the present disclosure.

[0033] like Figure 2 As shown, the big model-based comment display method 200 may include operations S210 to S230.

[0034] In operation S210 , in response to determining that the target resource is a hotspot resource, a target style matching the target resource is determined.

[0035] For example, hot resources refer to resources with high popularity, and whether the target resource is a hot resource can be determined by popularity parameters such as click volume and page views.

[0036] For example, the correspondence between resources and styles can be pre-configured. For example, the style corresponding to social news resources is controversial, and the style corresponding to chicken soup resources is encouraging. For another example, the target style of the target resource can be determined based on the big model. For example, the attribute information of the target resource and the prompt information (prompt) used to guide the big model to determine the target style are input into the big model. The prompt information can provide style options and guide the big model to select a suitable style as the target style. The prompt information can be pre-configured, and this embodiment does not limit the prompt information.

[0037] In operation S220 , based on the large model and according to the attribute information of the target resource and the target style, a target review targeting the target resource and conforming to the target style is generated.

[0038] For example, the target resource may be a resource of a data type such as video, image, audio, or graphic. For example, the attribute information of the target resource includes at least one of the following: title, text, tag, subtitle, content description text, and audio. It should be noted that the attribute information is determined based on the target resource. If the target resource is uploaded by the user, the acquisition and use of the attribute information are known and agreed to by the user, and are in compliance with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0039] For example, the target style and the attribute information of the target resource may be combined with the prompt information for guiding the large model to generate comments, and then input into the large model, so that the large model outputs the target comments that conform to the target style according to the attribute information of the target resource.

[0040] In operation S230 , the target comment is displayed.

[0041] For example, the target comment may be displayed in a comment area associated with the target resource.

[0042] The technical solution provided by the disclosed embodiment first determines the hot resources from the target resources, then determines the matching target style for these target resources, and then generates target comments based on the large model according to the attribute information and target style of the target resources. First, for different target resources, personalized comments can be generated from the content dimension and style dimension, so as to avoid the single style and content of the comments. At the same time, the large model can simulate human language habits and expressions, and the generated comments are of high quality, thereby improving the match between the comments and user needs, further improving the accuracy of the comments, and improving the user experience. Secondly, this method can automatically generate comments for each target resource, thereby reducing the complexity of user operations and improving the efficiency of generating target comments.

[0043] According to another embodiment of the present disclosure, the target style that matches the target resource can be determined in the following manner: the candidate style corresponding to the target category of the target resource can be determined as the target style according to the target category of the target resource and the correspondence between the category and the candidate style. For example, the correspondence between the resource and the candidate style can be pre-configured. For example, the style corresponding to social news resources is controversial, and the style corresponding to chicken soup resources is encouraging. This embodiment pre-configures the correspondence between the candidate style and the category, and the correspondence ensures the matching of the target style selection and avoids the low matching between the target resource and the target style. For example, the target resource and the target style do not match: the target style corresponding to the comedy resource is suspense. Obviously, the comedy resource has a higher matching with the funny style, but a lower matching with the suspense style. In addition, according to the target category and the pre-configured correspondence, the target style can be accurately determined, and at the same time, the style determination process is simplified, and the efficiency of determining the target style is improved.

[0044] According to another embodiment of the present disclosure, whether the target resource is a hot resource can be determined in the following manner: it can be determined whether the heat parameter of the target resource within a predetermined time period is greater than or equal to a heat threshold corresponding to the heat parameter. If so, the target resource can be determined to be a hot resource, otherwise the target resource is determined to be a non-hot resource. For example, the heat parameter includes at least one of the following: click volume, pageview volume, comment volume, like volume, favorite volume, and share volume, and the heat threshold corresponds to the heat parameter one by one.

[0045] For example, if it is detected that the target resource satisfies at least one of the following conditions within a predetermined period of time, it can be determined that the target resource is a popular resource: the number of clicks on the target resource is greater than or equal to a click threshold, the number of views on the target resource is greater than or equal to a view threshold, the number of comments on the target resource is greater than or equal to a comment threshold, the number of likes on the target resource is greater than or equal to a like threshold, the number of collections on the target resource is greater than or equal to a collection threshold, and the number of shares on the target resource is greater than or equal to a share threshold.

[0046] This embodiment determines whether the target resource is a hotspot resource by comparing the size relationship between the heat parameter of the target resource and the heat threshold within a certain period of time, so that the hotspot resource can be accurately identified. In addition, based on the heat parameters of multiple dimensions, the heat situation of the target resource can be comprehensively evaluated, further improving the accuracy of determining the hotspot resource.

[0047] According to another embodiment of the present disclosure, a large model for generating target comments is obtained by fine-tuning using multiple training samples.

[0048] In one example, the training sample includes a review sample that matches the resource sample. Next, the process of determining the review sample is described.

[0049] For example, a subset whose popularity meets a predetermined condition can be determined from the comment set of the resource sample, and the subset includes at least one comment. For example, the predetermined condition can include at least one of the following: the number of likes of the comment is greater than or equal to a first quantity threshold, the number of replies to the comment is greater than or equal to a second quantity threshold, and the comments with higher popularity are filtered by the predetermined condition, and the comments with higher popularity can be added to the subset.

[0050] Next, the quality of each comment in the subset can be evaluated based on the large model to obtain a quality value. During the evaluation process, the quality of the comment can be evaluated by evaluating the richness of the content, professionalism, whether there are independent opinions, etc. In addition, each dimension can correspond to a sub-quality value, and the quality value can be determined by the weighted results of each sub-quality value. For example, the comment and the prompt information used to guide the large model to perform the evaluation can be input into the large model, and the large model outputs the quality value. The prompt information is, for example, "The content of the comment is XX, and the comment is scored from the dimensions of richness of the content, professionalism, whether there are independent opinions, etc."

[0051] Next, the review samples can be determined from the subset according to the quality value. For example, the reviews with quality values ​​greater than or equal to the quality threshold are added to the value subset, or a predetermined number of reviews with high quality values ​​are added to the value subset. The reviews in the subset can be used as review samples.

[0052] In addition, after obtaining the subset, the comments in the subset can be displayed, and then the evaluator's operations on each comment in the subset can be obtained, and then the subset can be filtered again based on the evaluator's operations, and the comments in the filtered subset can be used as comment samples.

[0053] This example determines whether the popularity of comments meets the predetermined conditions through dimensions such as likes and replies, and selects comments with higher popularity from multiple comments. In this way, the selected comments have a wider audience base and higher user participation, making the comment samples more representative. In addition, using a large model to evaluate the quality of comments in multiple dimensions can make the selected comment samples have higher quality.

[0054] In another example, the training sample also includes the style of the comment sample. Next, the process of determining the style of the comment sample is described.

[0055] After obtaining a subset including at least one comment sample, the style of the comment sample can be annotated. For example, the classification prompt information template can be used to guide the large model to determine the style of the comment sample from multiple styles. For example, the classification prompt information template and the comment sample can be input into the large model, and the large model can output the style of the comment sample.

[0056] In some embodiments, the classification prompt information template may include examples, including comment examples and the style of the comment examples. In addition, the classification prompt information template may also provide style options, such as "funny", "controversial", "encouraging", etc., and guide the big model to select the style of the comment sample from various options.

[0057] This example uses the classification prompt information template to guide the big model to annotate the comment style. Using the capabilities of the big model for annotation can improve the accuracy and consistency of style recognition. The examples and clear style options in the classification prompt information template provide clear annotation guidance for the big model, improving the reliability of the annotation results.

[0058] In another example, the training sample includes sample attributes of the resource sample, a review sample matching the resource sample, a style of the review sample, and a training prompt information template.

[0059] For example, the sample attributes of the resource sample are similar to the attribute information of the target resource, for example, the sample attributes include at least one of the following: title, text, label, subtitle, content description text and audio. Taking the style of the comment sample as "funny" as an example, the training prompt information template is, for example, "the title of the comment sample is AA, the original text is BB, the subtitle is CC, generate a funny style comment".

[0060] For example, the training samples may include corresponding training prompt information and comment samples, wherein the training prompt information is obtained by combining the training prompt information template, the sample attributes of the resource sample, and the style of the comment sample.

[0061] Figure 3 It is a schematic diagram of the training process of a large model according to an embodiment of the present disclosure.

[0062] In this embodiment, the comment set 302 of the resource sample 301 can be screened based on popularity, and some comments with higher popularity can be screened from the comment set 302 and added to the subset 303. Then, the quality of each comment in the subset 303 can be evaluated to obtain a quality value, and then the comment set 302 can be screened based on the quality value, and some comments with higher quality values ​​can be screened from the subset 303 as comment samples 304.

[0063] In addition, the large model 311 can be guided to determine the style 307 of the comment sample 304 from multiple styles based on the classification prompt information template. The sample attribute 305 of the resource sample 301 can also be determined, and then the sample attribute 305 of the resource sample 301, the style 307 of the comment sample 304 and the training prompt information template 306 are combined into training prompt information 308. Then, the comment sample 304 and the training prompt information 308 are constructed into a training sample 309.

[0064] After generating the training samples 309, the training samples 309 can be used to train the large model 311. For example, a suitable base model 310 can be selected first, and then the base model can be fine-tuned by SFT (Sequence-to-Sequence Fine-Tuning) using the training samples 309. The base model 310 can be deeply trained and optimized to make the comments generated by it more natural, diverse, and closely fit the content of the material. The trained base model 310 can be used as the large model 311 for generating the target comments. This embodiment does not limit the fine-tuning process.

[0065] Figure 4 It is a schematic diagram of the comment display method according to an embodiment of the present disclosure.

[0066] After obtaining the trained big model, the big model can be used to generate target comments 405. For example, the target resource can be obtained first, and then the heat parameter of the target resource within a predetermined time period is compared with the heat threshold corresponding to the heat parameter to determine whether the target resource is a hot resource, wherein the heat parameter includes at least one of the following: click volume, pageview volume, comment volume, like volume, collection volume, and share volume. If the target resource is not a hot resource, the next target resource can be re-acquired, and it is again determined whether the newly acquired target resource is a hot resource. If the target resource is a hot resource, the candidate style corresponding to the target category 401 of the target resource can be determined as the target style 402 according to the target category 401 of the target resource and the correspondence between the category and the candidate style. Next, the attribute information 403 of the target resource, the target style 402 and the prompt information template 404 can be input into the big model, and the big model generates a target comment 405 for the target resource and conforms to the target style 402, and displays the target comment 405.

[0067] According to another embodiment of the present disclosure, historical comments of the target resource can also be obtained. For example, for a certain target resource, the comments that have been displayed in the comment area associated with the target resource are called historical comments. It should be noted that the acquisition and use of historical comments are known and agreed to by the users, and are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals. After obtaining the historical comments, you can first determine whether the historical comments are interrogative sentences. If they are not interrogative sentences, you no longer need to determine whether the historical comments represent object requirements. If they are interrogative sentences, you can determine whether the historical comments represent object requirements based on the large model.

[0068] After determining that the historical comments are interrogative sentences, the historical comments and the prompt information used to guide the big model to determine whether the comments contain clear questions can be input into the big model, and the big model determines whether the historical comments represent object requirements. It should be noted that some interrogative sentences contain user requirements, such as "What is the background music of the video", but some interrogative sentences do not contain object requirements. For example, one comment is "Waiting in line for 7 hours" and the other comment is "Only 7 hours?" Although the second reply comment is an interrogative sentence, it does not contain user requirements.

[0069] This embodiment first determines whether the historical comment is a question sentence, and if the historical comment is a question sentence, further determines whether the historical comment represents the object demand, rather than determining whether all historical comments represent the object demand. This allows for preliminary screening, reduces the number of calls to the large model, and thus improves processing efficiency and saves computing resources. In addition, using the ability of the large model to determine whether the comment contains a clear question improves the accuracy and efficiency of demand identification, and does not require developers to design complex algorithms, reducing the difficulty of development.

[0070] It is understandable that in other embodiments, it is not necessary to determine whether a historical comment is a question sentence, but rather all historical comments are processed to determine whether they represent an object requirement.

[0071] In one example, in the process of determining whether a historical comment is a question sentence, it is possible to detect whether the historical comment contains key characters, such as "?", "what", "what", etc. If the historical comment contains the key characters, it can be determined that the historical comment is a question sentence, otherwise it is determined that the historical comment is not a question sentence. This embodiment determines whether the historical comment is a question sentence by detecting representative key characters. This method can quickly identify question sentences and improve processing efficiency.

[0072] In another example, in the process of determining whether a historical comment is a question sentence, a classification model may be used to determine whether the historical comment is a question sentence. The classification model is pre-trained, and this embodiment does not limit the classification model and its training process.

[0073] After detecting that the historical comments of the target resource represent the object demand, first response data for the historical comments can be generated based on the big model according to the historical comments and the attribute information of the target resource, and then the first response data can be displayed.

[0074] For example, historical comments, attribute information of resources, and prompt information for guiding the big model to answer questions are input into the big model, and the big model outputs the first answer data to answer the user's questions or meet other requirements raised by the user. The big model can be ERNIE 4.0 Turbo or other models, and this embodiment does not limit the big model.

[0075] This embodiment automatically detects whether historical comments contain object requirements, and when there is a requirement, it uses the capabilities of the big model to accurately understand the user's needs based on the content of the historical comments and the attribute information of the target resource, and then generates first response data that meets the user's needs and automatically replies, thereby providing users with high-quality answers and improving the user experience.

[0076] According to another embodiment of the present disclosure, historical comments of the target resource may also be obtained. For example, for a certain target resource, the comments displayed in the comment area associated with the target resource are called historical comments. Then, it may be detected whether the historical comments include a predetermined wake-up character. The predetermined wake-up character may be pre-configured, for example, the predetermined wake-up character is "@", "question", "please answer", "intelligent assistant", etc.

[0077] If it is detected that the historical comments of the target resource do not include the predetermined wake-up character, the wake-up logic may not be triggered.

[0078] If it is detected that the historical comments of the target resource include a predetermined wake-up character, it means that the user wants to wake up the agent, so the wake-up logic can be triggered. The wake-up logic is explained next.

[0079] In one example, if it is detected that historical comments of the target resource include a predetermined wake-up character and text to be processed, based on the large model, second response data for answering the text to be processed is generated according to attribute information of the text to be processed and the target resource, and the second response data is displayed.

[0080] For example, a historical comment is "@Smart Assistant, where are the scenic spots in the video", where the predetermined wake-up characters include "@Smart Assistant", and the text to be processed includes "where are the scenic spots in the video". The text to be processed, the attribute information of the target resource, and the prompt information used to guide the large model to generate answers can be input into the large model, and the large model outputs the second answer data to answer the user's question. The second answer data may include data of the types of text, emoticons, images, animated images, audio, etc. In actual applications, the text content in the second answer data can be displayed below the historical comments, or multimedia content such as pictures and links in the second answer data can be sent to the user through the message center. Alternatively, all the second answer data can be displayed below the historical comments.

[0081] This example intelligently identifies the predetermined wake-up characters and text to be processed in historical comments, combines the attribute information of the target resource, and uses a large model to generate the second response data. Users only need to comment on the target resource and ask their own questions to get answers, without having to query the answers through search engines or other programs. This reduces the operation steps and improves the user experience.

[0082] In another example, if it is detected that the historical comments of the target resource include a predetermined wake-up character and lack text to be processed, interaction data for interacting with the object is determined, and then the interaction data is displayed.

[0083] For example, if the historical comment is "@intelligent assistant", and the user has not entered any text to be processed to indicate a clear need, the interaction data can be displayed. For example, the interaction data can be text, emoticons, etc. used to greet the user, or the interaction data can be the self-introduction of the intelligent assistant. In actual applications, the interaction data can be pre-configured or generated based on a large model. This embodiment does not limit the method for obtaining the interaction data.

[0084] In this embodiment, for the case where historical comments include a predetermined wake-up character and lack text to be processed, even if the user does not explicitly express his needs, a friendly greeting or self-introduction can be proactively provided. Compared to the method of not replying to the user, the interactive text in this embodiment can give the user feedback after the user enters the predetermined wake-up character, thereby improving the accuracy of the reply and thus improving the user experience.

[0085] Figure 5 It is a schematic structural block diagram of a comment display device according to an embodiment of the present disclosure.

[0086] like Figure 5 As shown, the large model-based comment display device 500 may include a style determination module 510 , a first generation module 520 and a first display module 530 .

[0087] The style determination module 510 is used to determine a target style that matches the target resource in response to determining that the target resource is a hotspot resource.

[0088] The first generating module 520 is used to generate a target comment for the target resource and conforming to the target style based on the large model and according to the attribute information and target style of the target resource.

[0089] The first display module 530 is used to display the target comment.

[0090] According to another embodiment of the present disclosure, the style determination module includes: a style determination submodule, which is used to determine a candidate style corresponding to a target category of the target resource as a target style according to a target category of the target resource and a correspondence between the category and multiple styles.

[0091] According to another embodiment of the present disclosure, the device also includes: a first determination module, for determining that the target resource is a hot resource in response to determining that the heat parameter of the target resource within a predetermined time period is greater than or equal to a heat threshold corresponding to the heat parameter; wherein the heat parameter includes at least one of the following: click volume, page views, comment volume, like volume, favorite volume, and share volume.

[0092] According to another embodiment of the present disclosure, a large model for generating target comments is obtained by fine-tuning using multiple training samples, each training sample including: sample attributes of a resource sample, a comment sample matching the resource sample, the style of the comment sample, and a training prompt information template.

[0093] According to another embodiment of the present disclosure, the comment sample is obtained through the following modules: a subset determination module, an evaluation module and a comment sample determination module. The subset determination module is used to determine a subset whose popularity meets a predetermined condition from the comment set of the resource sample, and the subset includes at least one comment. The evaluation module is used to evaluate the quality of each comment in the subset based on the large model to obtain a quality value. The comment sample determination module is used to determine the comment sample from the subset according to the quality value.

[0094] According to another embodiment of the present disclosure, the style of the comment sample is generated by the following modules: a comment style module, which is used to guide the large model to determine the style of the comment sample from multiple styles based on the classification prompt information template. The classification prompt information template includes examples, and the examples include comment examples and the style of the comment examples.

[0095] According to another embodiment of the present disclosure, the device further includes: a second generation module and a second display module. The second generation module is used to generate first response data for the historical comments based on the large model, according to the historical comments and the attribute information of the target resource in response to detecting the historical comments representing the object requirement of the target resource. The second display module is used to display the first response data.

[0096] According to another embodiment of the present disclosure, the apparatus further includes: a second determination module for determining, in response to determining that the historical comment is a question sentence, whether the historical comment represents an object demand based on the large model.

[0097] According to another embodiment of the present disclosure, the apparatus further includes: a third determination module, configured to determine that the historical comment is a question sentence in response to detecting that the historical comment includes a key character.

[0098] According to another embodiment of the present disclosure, the device further includes: a third generation module and a third display module. The third generation module is used to generate second response data for answering the text to be processed based on the large model and attribute information of the text to be processed and the target resource in response to detecting that the historical comments of the target resource include a predetermined wake-up character and the text to be processed. The third display module is used to display the second response data.

[0099] According to another embodiment of the present disclosure, the device further includes: an interaction data determination module and a fourth display module. The interaction data determination module is used to determine the interaction data for interacting with the object in response to detecting that the historical comments of the target resource include a predetermined wake-up character and lack the text to be processed. The fourth display module is used to display the interaction data.

[0100] According to another embodiment of the present disclosure, the attribute information includes at least one of the following: a title, a body, a tag, a subtitle, a content description text, and an audio.

[0101] Figure 6 The structural block diagram of an intelligent agent according to an embodiment of the present disclosure is schematically shown.

[0102] In the embodiments of the present disclosure, inspired by the von Neumann structure in modern computer theory, such as Figure 6 As shown, the agent 600 may include multiple core modules: an input module 610 , a processing module 620 and an output module 630 .

[0103] In the example, the input module 610 is responsible for receiving or sensing information such as queries, requests, instructions, signals, problem information or data from the outside world (such as users or external environment), and converting it into a format that the agent 600 can understand and process. The input module 610 is the primary link for the agent 600 to interact with the outside world. It enables the agent 600 to efficiently and accurately obtain the necessary "sensory" information from the outside world and respond to this information. In the example, the input module 610 can be used to receive input information. The input information may include attribute information of the target resource, historical comments, etc. as described above.

[0104] In the example, the processing module 620 is the core support for the ability of the agent 600 to handle complex tasks. The processing module 620 is used to determine the target task based on the input information received by the input module 610. The large model is determined based on the target task. By calling the large model, the comment display method based on the large model described above can be executed to obtain output information.

[0105] In an example, the output module 630 may be used to output the output information obtained by the processing module 620. The output information may include the target comments, the first response data, the second response data, and the interaction data as described above.

[0106] In an example, the processing module 620 may include a control unit 621 , a storage unit 622 , and an operation unit 623 .

[0107] During operation, the control unit 621 will continuously interact with the storage unit 622, the computing unit 623 and / or the output module 630. However, it should be noted that in the embodiment of the present disclosure, the control unit 621 acts as a single initiator to initiate communication with the storage unit 630, the computing unit 623 and / or the output module 630, and there is no communication coupling between the storage unit 622, the computing unit 623 and the output module 630.

[0108] In the example, the performance of the control unit 621 may be closely related to the large model based on the agent 600. In order to give full play to the capabilities of the large language model, the internal structure of the control unit 621 may be designed to be highly configurable and scalable to cope with various types of tasks and requirements in real scenarios.

[0109] The storage unit 622 may be responsible for memorizing information such as historical conversations, event flows, etc. Various prompt information templates, target styles, etc. as described above may be included in the storage unit 622 .

[0110] In the example, after the agent 600 obtains the input information, the agent 600 can use the input information to determine the target task. The target style is determined based on the target task. The target style can be stored in the storage unit 622. The agent 600 can retrieve the relevant prompt information template and target style from the storage unit 622 and feed it back to the control unit 621. Then, the control unit 621 can use the feedback prompt information template and target style to determine the big model based on the target task. The target comment for the target resource is obtained by calling the big model, and the target comment is passed to the output module 630.

[0111] The operation unit 623 can be regarded as a predefined tool library, such as a search engine, a format conversion tool, etc., which can be included in the operation unit 623 .

[0112] In the example, when the AI ​​agent 600 needs to render multiple output data, the relevant renderer and display control can be called from the operation unit 623 and fed back to the processing module 620. Then, the processing module 620 can use the feedback renderer and display control to pass the rendered result to the output module 630. It can be understood that although the large language model has excellent language understanding and generation capabilities, it is the same as a human being. Without the help of any tools, the tasks that can be solved are very limited. When the AI ​​agent 600 is given the ability to call tools, it can achieve tasks such as completing mathematical operations with the help of a calculator, completing data analysis with the help of python, and completing weather forecasts with the help of search engines.

[0113] According to the embodiment of the present disclosure, the intelligent agent 600 can simply and effectively improve the intelligence level, and improve flexibility and versatility.

[0114] In practical applications, intelligent agents can be applied in a variety of scenarios.

[0115] For example, it can be applied to the comment section of an information stream to automatically generate comments in the comment section and answer users' questions.

[0116] For example, when applied to social media platforms, it can actively participate in topic discussions, generate interesting comments, and attract more users to interact. It can also identify and respond to users' questions or comments and provide personalized recommendations or suggestions.

[0117] For example, when applied to e-commerce platforms, comments on product features can be generated on the product details page or purchase evaluation area to help users understand product details. The system can also respond to users' purchase inquiries, provide size recommendations, matching suggestions, etc., to enhance users' shopping experience.

[0118] For example, when applied to online education platforms, comments such as learning experiences and course summaries can be generated in the course comment area to create a learning atmosphere. It can also answer users' learning questions and provide recommendations for learning materials or learning methods.

[0119] For example, when applied to news websites or applications, comments related to the news content can be generated in the news comment area to guide users to conduct in-depth discussions. It can also identify and respond to users' questions or opinions on news events, and provide fact-checking or answers from multiple perspectives.

[0120] For example, when applied to entertainment applications, interesting comments related to the content can be generated in the comment area of ​​music, video and other entertainment applications to enhance the user's entertainment experience. It can also reply to users' comments or questions about songs, movies and other content, provide similar recommendations or share the stories behind them.

[0121] For example, it can be applied to smart assistants and serve as users' daily assistants, providing a unified interactive interface in various applications to help users manage tasks and obtain information.

[0122] For example, when used as a virtual partner, users can communicate with each other emotionally, share daily life details, and provide emotional support in chat applications.

[0123] For example, when applied to enterprise customer service scenarios, it can act as an intelligent customer service to promptly respond to users' inquiries and questions and provide uninterrupted service support. It can also identify user needs and emotions, and the intelligent agent can provide personalized solutions or soothing measures to improve customer satisfaction.

[0124] Applied to the above scenarios, it can provide users with more convenient and efficient review content, encourage users to actively participate in consultation discussions, enhance the interactivity of the application, and improve user experience.

[0125] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, including at least one processor; and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned comment display method.

[0126] According to an embodiment of the present disclosure, the present disclosure further provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to execute the above-mentioned comment display method.

[0127] According to an embodiment of the present disclosure, the present disclosure further provides a computer program product, including a computer program, and the computer program implements the above-mentioned comment display method when executed by a processor.

[0128] Figure 7 1 is a block diagram of an electronic device for implementing the comment display method of an embodiment of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0129] like Figure 7 As shown, the device 700 includes a computing unit 701, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 702 or a computer program loaded from a storage unit 708 into a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the device 700 can also be stored. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0130] A number of components in the device 700 are connected to the I / O interface 705, including: an input unit 706, such as a keyboard, a mouse, etc.; an output unit 707, such as various types of displays, speakers, etc.; a storage unit 708, such as a disk, an optical disk, etc.; and a communication unit 709, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 709 allows the device 700 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0131] The computing unit 701 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 701 performs the various methods and processes described above, such as the comment display method. For example, in some embodiments, the comment display method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 708. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded into the RAM 703 and executed by the computing unit 701, one or more steps of the comment display method described above may be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to perform the comment display method in any other appropriate manner (e.g., by means of firmware).

[0132] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0133] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0134] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0135] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0136] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0137] A computer system may include clients and servers. Clients and servers are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship to each other.

[0138] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0139] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A comment display method based on a large model, comprising: In response to determining that the target resource is a hotspot resource, determining a target style that matches the target resource; Based on the big model, according to the attribute information of the target resource and the target style, generate a target comment for the target resource and in accordance with the target style; as well as The target review is displayed.

2. The method according to claim 1, wherein: Determining a target style that matches the target resource includes: According to the target category of the target resource and the correspondence between the category and the candidate style, the candidate style corresponding to the target category is determined as the target style.

3. The method according to claim 1, further comprising: In response to determining that a heat parameter of the target resource within a predetermined time period is greater than or equal to a heat threshold corresponding to the heat parameter, determining that the target resource is the hotspot resource; The popularity parameters include at least one of the following: clicks, views, comments, likes, favorites, and shares.

4. The method according to claim 1, wherein: The large model used to generate the target comment is obtained by fine-tuning using multiple training samples, each of which includes: sample attributes of a resource sample, a comment sample matching the resource sample, the style of the comment sample, and a training prompt information template.

5. The method according to claim 4, wherein: The review samples were obtained in the following way: Determining a subset whose popularity meets a predetermined condition from the comment set of the resource sample, wherein the subset includes at least one comment; Evaluate the quality of each review in the subset based on the large model to obtain a quality value; as well as The review sample is determined from the subset according to the quality value.

6. The method according to claim 4, wherein: The style of the review sample is generated in the following way: Guiding the large model to determine the style of the review sample from multiple styles based on the classification prompt information template; The classification prompt information template includes examples, and the examples include comment examples and the style of the comment examples.

7. The method according to any one of claims 1 to 6, further comprising: In response to detecting that the historical reviews of the target resource represent an object requirement, generating first response data for the historical reviews based on the big model and according to the historical reviews and attribute information of the target resource; as well as The first response data is displayed.

8. The method according to claim 7, further comprising: In response to determining that the historical review is a question sentence, it is determined based on a large model whether the historical review represents an object requirement.

9. The method according to claim 8, further comprising: In response to detecting that the historical comment includes a key character, determining that the historical comment is a question sentence.

10. The method according to claim 1, further comprising: In response to detecting that the historical comments of the target resource include a predetermined wake-up character and a text to be processed, generating second answer data for answering the text to be processed based on the large model and according to the attribute information of the text to be processed and the target resource; as well as The second response data is displayed.

11. The method according to claim 9, wherein: The method further comprises: In response to detecting that the historical comments of the target resource include a predetermined wake-up character and lack text to be processed, determining interaction data for interacting with the object; and The interaction data is displayed.

12. The method according to claim 1, wherein: The attribute information includes at least one of the following: title, body, tag, subtitle, content description text, and audio.

13. A review display device based on a large model, comprising: A style determination module, configured to determine a target style matching the target resource in response to determining that the target resource is a hotspot resource; A first generating module is used to generate a target comment for the target resource and conforming to the target style based on the large model and according to the attribute information of the target resource and the target style; as well as The first display module is used to display the target comment.

14. The device according to claim 13, wherein: The style determination module includes: The style determination submodule is used to determine the candidate style corresponding to the target category as the target style according to the target category of the target resource and the correspondence between the category and the candidate style.

15. The apparatus according to claim 13, further comprising: A first determination module, configured to determine that the target resource is the hotspot resource in response to determining that a heat parameter of the target resource within a predetermined time period is greater than or equal to a heat threshold corresponding to the heat parameter; The popularity parameters include at least one of the following: clicks, views, comments, likes, favorites, and shares.

16. The device according to claim 13, wherein: The large model used to generate the target comment is obtained by fine-tuning using multiple training samples, each of which includes: sample attributes of a resource sample, a comment sample matching the resource sample, the style of the comment sample, and a training prompt information template.

17. The device according to claim 16, wherein: The review samples are obtained through the following modules: A subset determination module, used to determine a subset whose popularity meets a predetermined condition from the comment set of the resource sample, wherein the subset includes at least one comment; An evaluation module, used for evaluating the quality of each review in the subset based on the large model to obtain a quality value; as well as The review sample determination module is used to determine the review sample from the subset according to the quality value.

18. The device according to claim 16, wherein: The style of the review sample is generated through the following modules: A comment style module, used to guide the large model to determine the style of the comment sample from multiple styles based on the classification prompt information template; The classification prompt information template includes examples, and the examples include comment examples and the style of the comment examples.

19. The device according to any one of claims 13 to 18, further comprising: A second generating module is used for generating first response data for the historical review based on the big model and according to the historical review and the attribute information of the target resource in response to detecting the historical review representation object requirement of the target resource; as well as The second display module is used to display the first response data.

20. The apparatus according to claim 19, further comprising: The second determination module is used to determine whether the historical review represents an object demand based on the large model in response to determining that the historical review is a question sentence.

21. The apparatus according to claim 20, further comprising: The third determination module is configured to determine that the historical comment is a question sentence in response to detecting that the historical comment includes a key character.

22. The apparatus according to claim 13, further comprising: A third generating module is used for generating second answer data for answering the text to be processed based on the large model and attribute information of the text to be processed and the target resource in response to detecting that the historical comments of the target resource include a predetermined wake-up character and the text to be processed; as well as The third display module is used to display the second response data.

23. The apparatus according to claim 13, further comprising: an interaction data determination module, configured to determine interaction data for interacting with an object in response to detecting that the historical comments of the target resource include a predetermined wake-up character and lack text to be processed; as well as The fourth display module is used to display the interaction data.

24. The device according to claim 13, wherein: The attribute information includes at least one of the following: title, body, tag, subtitle, content description text, and audio.

25. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 12.

26. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 12.

27. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1 to 12.

28. An intelligent agent, comprising: An input module, used for receiving input information; A processing module, configured to obtain output information by calling a large model to execute the method according to any one of claims 1 to 12 based on the input information received by the input module; An output module is used to output the output information obtained by the processing module.

Images