Comment processing method and device, equipment and storage medium

Abstract

The invention provides a comment processing method and device, equipment and a storage medium, and relates to the technical field of artificial intelligence, in particular to the technical fields of large models, data screening, information processing, intelligent recommendation and the like. According to the specific implementation scheme, at least two candidate comments of a target resource are generated by adopting a first large model; screening a first recommendation comment corresponding to the target object from the at least two candidate comments; and in response to the comment acquisition request, recommending the first recommended comment to the target object.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to technical fields such as large models, data screening, information processing, and intelligent recommendation. Background Art

[0002] With the rapid development of Internet technology and social media, the ways in which users obtain information have become more diverse and instant, which not only increases the accessibility of information but also promotes the interaction between users and content.

[0003] Comments, as an important form of interaction, have become a bridge connecting users and content. They not only reflect users' views and feedback on specific content, but also provide a reference perspective for other users. Summary of the invention

[0004] The present disclosure provides a comment processing method, apparatus, device and storage medium.

[0005] According to one aspect of the present disclosure, a comment processing method is provided, comprising:

[0006] Using the first model to generate at least two candidate reviews of the target resource;

[0007] Filtering a first recommended comment corresponding to the target object from at least two candidate comments;

[0008] In response to the comment acquisition request, a first recommended comment is recommended to the target object.

[0009] According to one aspect of the present disclosure, a comment processing method is provided, comprising:

[0010] Obtaining a first recommended comment of a target resource; the first recommended comment is generated based on a first large model and is selected for the target object from at least two candidate comments generated by the first large model;

[0011] When the target object's terminal device displays the target resource, the first recommended comment is displayed.

[0012] According to another aspect of the present disclosure, there is provided a comment processing device, comprising:

[0013] A generating module, configured to generate at least two candidate reviews of a target resource using the first large model;

[0014] A screening module, used to screen a first recommended review corresponding to a target object from at least two candidate reviews;

[0015] The recommendation module is used to recommend a first recommended comment to a target object in response to a comment acquisition request.

[0016] According to another aspect of the present disclosure, there is provided a comment processing device, comprising:

[0017] An acquisition module, configured to acquire a first recommended comment on a target resource; the first recommended comment is generated based on a first large model and is selected for the target object from at least two candidate comments generated by the first large model;

[0018] The display module is used to display the first recommended comment when the terminal device of the target object displays the target resource.

[0019] According to another aspect of the present disclosure, there is provided an electronic device, comprising:

[0020] at least one processor; and

[0021] a memory communicatively connected to the at least one processor; wherein,

[0022] 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 execute any method in the embodiments of the present disclosure.

[0023] 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 the computer to execute any method according to the embodiments of the present disclosure.

[0024] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, which implements any method according to the embodiments of the present disclosure when executed by a processor.

[0025] 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

[0026] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0027] Figure 1 is a flowchart of a comment processing method provided according to an embodiment of the present disclosure;

[0028] Figure 2 is a schematic diagram of a process for generating multiple candidate comments according to an embodiment of the present disclosure;

[0029] Figure 3 is a schematic diagram of a process for screening recommended comments according to an embodiment of the present disclosure;

[0030] Figure 4is a schematic diagram of a framework of a recall ranking model provided according to an embodiment of the present disclosure;

[0031] Figure 5 It is a schematic diagram of a process of generating candidate comments according to changes in popularity provided by an embodiment of the present disclosure;

[0032] Figure 6 is a flowchart of a comment recommendation method provided according to an embodiment of the present disclosure;

[0033] Figure 7 It is a schematic diagram of a process of generating comments in two modes provided according to an embodiment of the present disclosure;

[0034] Figure 8 is a schematic diagram of a recommendation comment display interface provided according to an embodiment of the present disclosure;

[0035] Fig. 9 is another schematic diagram of a recommendation comment display interface provided according to an embodiment of the present disclosure;

[0036] Fig.10 is a schematic diagram of another recommendation comment display interface provided according to an embodiment of the present disclosure;

[0037] Fig.11 It is a schematic diagram of the overall process of recommending comments according to an embodiment of the present disclosure;

[0038] Fig.12 is a structural diagram of a comment processing device provided according to an embodiment of the present disclosure;

[0039] Fig.13 is a structural schematic diagram of another comment processing device provided according to an embodiment of the present disclosure;

[0040] Fig.14 It is a block diagram of an electronic device used to implement the comment processing method according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0041] 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 of the present disclosure. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0042] The terms "first", "second", etc. in this disclosure are used to distinguish similar objects, and are not necessarily used to describe a specific order or precedence. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, including a series of steps or units. The method, system, product or device is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0043] With the rapid development of Internet technology and social media, users browse massive amounts of information on various information applications and social platforms every day. Comments are an important way for users to interact with information content, and their quality and efficiency directly affect the user experience.

[0044] However, faced with complex information, users may need to conceive and organize their own comments when writing them, which is not only time-consuming and laborious, but may also fail to fully express their views due to limited expressive ability. This not only reduces the enthusiasm of users to participate in the interaction, but also limits the diversity and depth of the comments, thus affecting the interactivity and user stickiness of the platform.

[0045] In view of this, an embodiment of the present disclosure provides a comment processing method, which is applicable to a server. The comment processing method can generate a rich plurality of candidate comments through a large model, and then filter out comments suitable for different target objects for recommendation, and display the recommended comments on the terminal device of the target object, which helps to simplify the comment publishing process and improve the comment publishing efficiency and user experience. Specifically, the comment processing method performed on the server side provided by the embodiment of the present disclosure mainly involves generating comments and filtering comments for different target objects. The following will be introduced from the perspective of a single target object. It can be understood that each target object is processed in the same way. Figure 1 As shown, it mainly includes the following contents:

[0046] S101, using a first large model to generate at least two candidate reviews for a target resource.

[0047] Target resources refer to content or platforms that target objects can browse, consume and comment on. These target resources usually involve multiple aspects such as goods, services, information, short videos, micro-films, live broadcasts, etc., and allow target objects to express their opinions and experiences in the form of comments.

[0048] The first large model usually refers to a pre-trained multimodal large model. A multimodal large model refers to a model that combines multimodal information such as text, images, audio, and video for processing. This model can process multiple types of data and achieve more efficient information response and intelligent recognition by fusing data from different modalities.

[0049] In the disclosed embodiment, the first large model can be pre-trained to meet the requirement of generating at least two candidate reviews for the target resource.

[0050] S102: Filter a first recommended comment corresponding to the target object from at least two candidate comments.

[0051] That is, a candidate comment that meets the interests and needs of the target object is selected from at least two candidate comments generated by the first model for the target resource, and is used as the first recommended comment recommended to the target object. The first recommended comment should be able to arouse the interest of the target object, have a certain consensus with the target object on the understanding of the target resource, and provide valuable information or opinions.

[0052] S103: In response to the comment acquisition request, recommend a first recommended comment to the target object.

[0053] In the disclosed embodiment, the resource recommendation engine can filter out multiple suitable resources from a large amount of resources for the target object and form a resource list to be pushed to the terminal device of the target object. Therefore, in a possible implementation, when the terminal device receives the resource list, the terminal device can send a comment acquisition request to the server, so that the server pushes the first recommended comments of all resources in the resource list to the terminal device, including the first recommended comment of the target resource.

[0054] In addition, segmented display and segmented recommended comments can also be supported. For example, the terminal device supports the display of n resources in the same display interface, where n is a positive integer greater than or equal to 1. If the number of resources in the recommendation list is greater than n, the resources in the recommendation list will be displayed in pages on the terminal device. Thus, when displaying the first page of resources, the terminal device can send a comment acquisition request for each resource in the first page to the server so that the server can push the first recommended comment. Before displaying the i-th page of resources after the first page (i is greater than 1), while displaying the iq-th page, the terminal device sends a comment acquisition request for the resources in the i-th page to the server, where q is a positive integer greater than or equal to 1, and when (iq) is less than 0, (iq) takes the value of 1, that is, a comment acquisition request is sent when the first page of resources is displayed.

[0055] In short, while the terminal device of the target object displays the detailed content of the target resource, the target object can simultaneously perceive the displayed first recommended comment. The embodiment of the present disclosure does not limit the triggering method and timing of the comment acquisition request.

[0056] In the disclosed embodiment, the first large model is used to generate at least two candidate comments for the target resource, and the powerful generation and reasoning capabilities of the large model can be used to provide a wealth of candidate comments for the target resource. These candidate comments are relatively consistent with the content of the target resource, and can give a description of the target resource from different perspectives or give content derived from the target resource. When recommending comment content for the target object, the candidate comment that meets the preferences of the target object can be screened out from at least two candidate comments for the target object as the first recommended comment, and the first recommended comment can be recommended to the target object when the target object browses the target resource. The first recommended comment is screened out based on the target object, thereby achieving accurate recommendation. At the same time, the voice organization ability of the first large model can also solve the problem of lengthy comments, simplify the complexity of operation, and improve user experience.

[0057] In the disclosed embodiment, the implementation method of using the first large model to generate at least two candidate reviews of the target resource is as follows: Figure 2 As shown, including:

[0058] S201, obtaining resource characteristics of a target resource.

[0059] That is, the required feature information is extracted from the target resource. These features include but are not limited to target resource classification information, target resource tag information, target resource release time, resource title, resource content, etc.

[0060] When the target resource is a multimodal resource, the characteristics of the target resource can be extracted from information of different modalities so that the first large model can fully understand the target resource.

[0061] During implementation, technologies such as OCR (Optical Character Recognition) and ASR (Automatic Speech Recognition) can be used to extract content understanding dimension features from images, videos, and audios in the target resources to convert them into a form that can be understood and processed by the first large model.

[0062] The information of the image modality can be processed by the visual processing module to understand the overall meaning of the image modality, the details of the image content and other information.

[0063] S202: Process resource features based on the first large model and at least two comment style templates to generate at least two candidate comments with at least two comment styles.

[0064] It can be understood that a prompt (prompt template) of a suitable style is selected according to the characteristics of the target resource, and multiple candidate comments with various comment styles are generated to enrich the comment candidate library, so as to facilitate the selection of the first recommended comment for the target object from the comment candidate library.

[0065] In the embodiment of the present disclosure, in order to ensure the diversity of generated comments, at least two prompt templates of comment styles may be preset, for example, the following templates may be included:

[0066] Controversy templates can make the generated comment sentences have distinct opinions;

[0067] Fun templates can generate comments that include popular Internet memes, jokes, and fun conversations.

[0068] Professional templates can make the generated comment sentences contain vertical professional knowledge;

[0069] The general class template can make the generated comment sentences have no obvious style tendency;

[0070] The deep template can make the generated comment sentences contain in-depth analysis and have the characteristics of extended thinking;

[0071] Encouragement templates can make the generated comments express encouragement in a positive way;

[0072] Story templates can generate comments that are stories related to the target resource or content that has been formed through multiple rounds of replies.

[0073] Positive energy templates can make the generated comment sentences clean, clear-cut, and promote mainstream values.

[0074] In the disclosed embodiment, different styles of initial prompt word templates can be preset, and high-quality comment sentences from outside or inside the site can be manually captured as training samples. The initial prompt word templates are continuously adjusted and optimized through the output results of the first large model to obtain prompt word templates of different styles that meet expectations. Taking the fun template as an example, the template content can be "Based on the title and summary, generate two humorous and funny comments. The comments can quote Internet buzzwords, hot topics, or use appropriate exaggerated language. Only generate comments without any additional information." The first large model is required to generate appropriate comment sentences based on the examples.

[0075] It is understandable that among the multiple candidate comments with various styles generated by the first model, each candidate comment has its own unique style while maintaining its association with the target resource, which can cover the preferences of different types of target objects. These candidate comments can be controversial, interesting, professional or general, in order to increase the diversity and richness of the comments, so as to provide personalized recommended comments for different target objects and improve the accuracy of the recommendation.

[0076] In the disclosed embodiment, by acquiring the resource characteristics of the target resource and generating candidate comments with a variety of comment styles based on the first large model and at least two comment style templates, the diversified comment styles can meet the needs of different target objects or scenarios, so as to facilitate the screening of the first recommended comments that meet the needs of various target objects for different target objects, thereby improving the accuracy of the recommendation.

[0077] In the embodiment of the present disclosure, the implementation method of selecting a first recommended comment corresponding to a target object from at least two candidate comments for the target object is as follows: Figure 3 As shown, including:

[0078] S301, sorting at least two candidate comments based on a first object feature of a target object and a resource feature of a target resource.

[0079] The first object feature is the attribute information of the first object.

[0080] During implementation, the first object feature of the target object and the resource feature of the target resource can be converted into a ranking index through the trained recall ranking model to measure the fit between the candidate review and the target object. The specific implementation steps are as follows:

[0081] Step A1, obtaining semantic features of at least two candidate comments;

[0082] As the name implies, it is information that can reflect the meaning extracted from multiple candidate comments. In implementation, semantic information can be extracted from the text data of multiple candidate comments by using a BGE (BAAI General Embedding, a semantic vector model) model to obtain the semantic features of each of the multiple candidate comments.

[0083] Step A2, inputting the first object feature, resource feature, and semantic feature of each candidate comment into a recall ranking model to obtain the target object's adoption probability of each candidate comment;

[0084] That is, the first object features, target resource features and semantic features of the candidate comments obtained in step A1 are integrated and input into the recall ranking model, and the target object's adoption probability of each candidate comment is obtained based on the recall ranking model. These probability values ​​reflect the degree of match between the candidate comment and the target object's interests and cognition of the target resource.

[0085] Step A3: sorting at least two candidate comments based on the adoption probability.

[0086] According to the target object's adoption probability of each candidate comment obtained in step A2, multiple candidate comments are sorted in descending order, and candidate comments with higher adoption probability are placed at the front. The sorting result will directly affect the comment list finally displayed to the target object, thereby affecting the target object's experience and satisfaction.

[0087] In the disclosed embodiment, by obtaining the semantic features of multiple candidate comments, the relationship between the core content expressed by the candidate comments and the target resource can be deeply understood. By inputting the first object feature of the target object, the resource feature, and the semantic features of each candidate comment into the recall ranking model, it is possible to comprehensively consider information from multiple dimensions and screen out the first recommended comment for the target object, thereby improving the accuracy of the recommendation for the target object and enhancing the user experience.

[0088] S302: Filter the first recommended comment based on the ranking of at least two candidate comments.

[0089] That is, by sorting the candidate comments, we can screen out the comments that best meet the needs and preferences of the target objects, thereby improving the efficiency of information processing and decision-making.

[0090] In the disclosed embodiment, by sorting multiple candidate comments based on the first object characteristics of the target object and the resource characteristics of the target resource, it is possible to consider from multiple dimensions so as to sort multiple candidate comments based on the characteristics of the target object, thereby accurately screening out the first recommended comments that are suitable for the target object, thereby improving the recommendation accuracy of the first recommended comments and improving the user experience.

[0091] The recall ranking model in the embodiment of the present disclosure can be obtained by collecting data of multiple sample objects as training samples and iteratively training the initial model. Figure 4As shown in the figure, the training samples are obtained by integrating the object features of the sample objects, the resource features of the sample resources, and the historical comment features. In the training samples, the features of these three dimensions are converted into vectors through the Embedding layer, and then the vectors are learned through multiple fully connected layers to predict the predicted adoption probability of the sample object for the sample comment. Based on the predicted adoption probability, it can be judged whether the sample object finally adopts the sample comment. The label of the training sample is used to describe whether the sample object actually adopts the sample comment. Therefore, the prediction results of the sample object for the sample comment are compared with the corresponding labels, and then the loss can be determined and the model parameters can be optimized. In summary, a recall ranking model that can accurately predict the adoption summary is obtained through supervised training, so as to screen out suitable recommended comments for different target objects. Since the first object features of different target objects are differentiated, different comments can also be screened out for different target objects from the same set of candidate comments.

[0092] Since the distribution effects of different target resources are different, the higher the click volume, the higher the popularity, the higher the possibility of the target object to post a comment. Therefore, in the face of different target resources, it is necessary to cater to the growth of the distribution scale of the target resources to expand the candidate library of candidate comments, so as to better recommend different comments for different target objects.

[0093] During implementation, when the target resource is released, the first model can be used to generate at least two candidate comments.

[0094] That is, when the target resource is just released or about to be released, the first trigger will request the first model to generate a certain number of candidate comments to meet the needs of different target objects for comments. During implementation, the number of candidate comments can be adjusted according to factors such as the target resource type and the expected level of interaction.

[0095] In the disclosed embodiment, when a target resource is detected to be released, at least two candidate comments are generated by using the first large model, and relevant candidate comments can be provided to the target object in a timely manner, so that the first recommended comment can be obtained in a timely manner when the target object flows to the target resource. By generating candidate comments by the first large model and screening the first recommended comments for the target object, the complexity of the operation of manually conceiving and writing the corresponding comment content can be reduced, thereby simplifying the process of the target object posting comments, improving the efficiency of comment posting, and improving the user experience.

[0096] In addition, when the target resource is released, the popularity change of the target resource can be obtained; when the value of the popularity change is a preset threshold, the first large model is used to generate at least two candidate comments.

[0097] In the disclosed embodiment, the target resource's popularity change indicators may include but are not limited to: the number of views of the target resource, the number of comments on the target resource, the number of likes or favorites of the target resource, the number of shares of the target resource, and the time spent on the target resource page.

[0098] In summary, Figure 5 For example, in S501, the target resource is published. In S502, the first large model is requested to generate at least two candidate comments. The generated candidate comments can be stored in the candidate library to facilitate the screening of the first recommended comment for the target object. In addition, in S503, the statistical start time of the first heat change can be determined. Afterwards, in S504, the heat change of the target resource starting from the statistical start time can be counted based on the statistical start time determined in S503. In S505, it is determined that the value of the heat change counted in S504 reaches the preset threshold, then S502 is triggered, and the first large model is continuously requested to generate at least two candidate comments to update the candidate library. At the same time, S503 is triggered to update the statistical start time of the hot spot change. Then, in S504, the heat change of the target resource is counted based on the new statistical start time to facilitate the subsequent process to be triggered again, and this loop execution can be performed until the target resource is removed from the shelves or the heat drops to the target heat.

[0099] For example, when the target resource is just released, the first model has generated n candidate comments for the target resource in real time. By real-time monitoring of the target resource's popularity change value, if the target resource's page views have increased by m and reached the first level, the first model will be called again to generate new candidate comments; and so on, when the target resource's page views have increased by m on the basis of the first level and reached the second level, the first model will be called again to generate new candidate comments, and so on, the candidate comments of the target resource are continuously supplemented.

[0100] By obtaining the value of the target resource's popularity change in real time and setting the corresponding preset threshold, candidate comments can be dynamically supplemented to provide different comments for different target objects. For example, after the target resource is released, the first largest model can be requested to add multiple additional candidate comments to the target resource every time the popularity of the subsequent target resource increases by m, so as to ensure that there are sufficient candidate comments for popular resources to recommend to different target objects. In addition, as the popularity of the resource changes, the threshold can be adjusted dynamically. For example, in the case where the popularity of the target resource is growing rapidly, the threshold can be appropriately lowered to generate candidate comments faster; conversely, in the case of slow growth in popularity, the threshold can be appropriately increased to avoid wasting the computing resources of the first largest model.

[0101] During implementation, in order to ensure the diversity of comments, deduplication operations will be performed on the comments generated in each round, so that the candidate comments generated in each round are different and will not be repeated with the candidate comments generated previously.

[0102] It is understandable that the first model has perturbation parameters, which will also cause differences between candidate comments generated for the same target resource. Even if the content is the same, the expression will be different. On this basis, further combining multiple comment style templates, the first model can continuously generate a variety of candidate comments for the target resource.

[0103] In the disclosed embodiment, by acquiring the popularity of the target resource, the popularity change of the target resource can be monitored in real time, and candidate comments can be supplemented in time to meet the needs of different target objects, thereby improving the accuracy and user experience of recommendations for different target objects. In addition, by setting a preset threshold, the generation of candidate comments can be dynamically adjusted, while ensuring the balance of the candidate comment generation speed, and also improving the utilization rate of the computing resources of the first large model.

[0104] For different target resources, the number of comments posted by different target objects varies. Some target objects actively participate in discussions and post more comments. In the same target resource, target objects that post more comments can be called high-rated users compared to target objects that post fewer comments.

[0105] Highly rated users are often more likely to post comments. At this point, in order to better help highly rated users post comments, in the disclosed embodiment, a combination of offline and online methods is used to recommend comments to highly rated users. It can be implemented as follows: based on the comments generated by the aforementioned first large model, not only are recommended comments screened out for highly rated users, but also, when the target resource is displayed and the target object belongs to a preset category (the preset category belongs to the highly rated user category), the second large model is used to process the second object features of the target object and the resource features of the target resource to generate a second recommended comment corresponding to the target object;

[0106] Then, the second recommended comment is recommended to the target object; wherein the first recommended comment and the second recommended comment are recommended to the target object in a predetermined order.

[0107] That is, when the preset category of the target object is a high-rated user, in addition to the candidate comment generation method described above, the second largest model will be simultaneously requested to generate a second recommended comment for the target object in real time in an online manner.

[0108] In the embodiment of the present disclosure, when it is determined that the target object browses the target resource and the target object belongs to a preset category, the second largest model is used to process the relevant features. Therefore, instead of generating the second recommended comments for all target objects, the second largest model is called to generate the second recommended comments for the target objects of the preset category, and the computing resources of the second largest model can be reasonably used to provide more comments for the target objects in need.

[0109] For example, after generating the second recommended comment online, since this type of recommended comment is tailored for the target object, at least one second recommended comment generated online can be inserted into the sequence of at least one first recommended comment generated by the first large model to be recommended to the target object.

[0110] For example, through the first recommended comment generated offline, the target object can perceive the first recommended comment at the first time when browsing the target resource, helping the target object to post a comment. By combining the online stream method, a personalized second recommended comment can be generated for the target object, so that the second recommended comment can be displayed first. During implementation, a comment display sequence can be constructed, and the first recommended comments generated by the first large model are displayed in order through the display sequence, and the second recommended comments generated by the second large model can be inserted into the display sequence for easy display. During implementation, since the second recommended comment is generated in real time, it can be required to display the second recommended comment first. For example, the first recommended comments generated offline include P1, P2, P3...Pn in sequence. When the second recommended comment is generated online, comments P1 and P2 are currently displayed, then the recommended comments generated online such as Z1 and Z1 can be inserted before P3 for display.

[0111] Among them, the target resource features include the target resource title, the target resource classification label, the content understanding information extracted from the pictures, videos and audio in the target resource using OCR, ASR and other technologies, etc. The second object feature may include the comment record of the target object. For example, on the basis of the first object feature, the comment record of the target object can be additionally added to capture the preferences and concerns expressed by the target object recently through the comment record, so as to timely perceive the recent changes of the target object in an online manner and generate personalized comment text for the target object that meets the characteristics of the target object.

[0112] During implementation, the recent comment history of the target object can be collected and stored at the daily granularity. When the target object is visited, the corresponding historical comment record will be called to provide data support for the large model, so that the model can obtain the target object's trends in real time, such as recent hot topics, catchphrases, and comment habits. In this way, through the comment record of the target object, the target object's interests can be more accurately grasped, so that the generated comments are more in line with the target object's preferences, so as to improve the generation accuracy of the second recommended comment and improve the user experience.

[0113] During implementation, when it is determined that the display time of the target resource by the terminal device of the target object reaches a specified time and the target object belongs to a preset category, the second largest model is used to process the second object characteristics and resource characteristics.

[0114] That is, when the target object is browsing the target resource, and the terminal device shows that the target object's stay time on the target resource reaches a specified duration, by acquiring the second object features and the target resource features in real time, at least one recommendation comment with personalized characteristics is generated for the target object based on the second largest model. Among them, the display duration of the target resource by the terminal device of the target object can reflect the target object's interest in the target resource. When the display duration reaches the specified duration, it indicates that the target object may have a strong interest in the target resource. In this case, by using the second largest model to process the second object features and resource features, a second recommendation comment with personal preferences can be generated for the target object in a timely manner. While improving the accuracy of the second recommendation comment, it can also make reasonable use of the resources of the second largest model and improve the rationality of resource utilization.

[0115] For example, calling the second largest model to generate comments in an online streaming manner may take at least 1 second, during which bandwidth and computing resources of the large model are required to generate comments online. When there are a large number of objects on the multimedia platform, if comments are expanded for each object by generating comments online, it will inevitably lead to resource squeeze and it will be difficult to meet the demand. At this time, by accurately locating the target objects of the preset category (i.e., high-rated users), and determining that they have a high willingness to post comments, the target objects with a strong willingness to post comments are selected from the numerous objects to generate comments for them online, so that computing resources can be reasonably used.

[0116] It can be understood that the first large model and the second large model can be the same model, and the first object feature and the second object feature can also be the same.

[0117] When the recommended comment generated by the first model is recommended to the target object and accepted by the target object, it will be deleted from the candidate comment library of the first model to avoid recommending the same comment to different target objects when making recommendations for other target objects.

[0118] In summary, the above article describes how to generate and screen candidate comments. In the embodiment of the present disclosure, based on the same technical concept, a comment processing method is also provided to illustrate how the terminal device processes the comments generated in the above article. Figure 6 As shown, the comment processing method may include the following contents:

[0119] S601, obtaining a first recommended comment on a target resource; the first recommended comment is generated based on a first large model, and is screened for the target object from at least two candidate comments generated by the first large model.

[0120] Based on what has been described above, the first recommended comment of the target resource obtained is screened from the candidate comments generated by the first model.

[0121] S602: When the target object's terminal device displays the target resource, display the first recommended comment.

[0122] That is, when it is detected that the terminal device of the target object displays the target resource, the first recommended comment can be displayed to the target object to assist the target object in posting a comment. In implementation, at least one first recommended comment of the target resource obtained in step S602 can be displayed to the target object in a variety of ways.

[0123] In the disclosed embodiment, the first recommended comment of the target resource is obtained. The recommended comment is often selected for the target object and can more accurately reflect the characteristics of the target resource and the needs of the target object, thereby accurately providing the first recommended comment for the target object. When the target object's terminal device displays the target resource, the first recommended comment is displayed so that the target object can observe the first recommended comment, eliminating the need for the target object to conceive and write comments, thereby simplifying the process of the target object posting comments and improving the user experience.

[0124] In the disclosed embodiment, on the one hand, the high resource cost of real-time request for large model to generate recommended comments can be reduced, while on the other hand, the target object can be responded to in a timely manner and a comment statement can be generated for the target object quickly. Figure 7 As shown, at least two first recommended comments generated offline for the target object by the first large model can be used. On the other hand, in order to ensure that a personalized comment can be generated for the target object by combining the target resource characteristics and the target object characteristics, a second recommended comment generated by the second large model combined with the second object characteristics of the target object can be obtained; and the second recommended comment is displayed in a predetermined order of the second recommended comment and the first recommended comment.

[0125] The second object feature is the attribute information of the second object. The second object feature is processed by the second largest model, and the second recommended comment with the target object preference can be generated online. The second recommended comment is displayed according to the predetermined order of the first recommended comment and the second recommended comment as described above.

[0126] In the disclosed embodiment, the second recommended comment generated by the second largest model combined with the second object feature of the target object makes the generated recommended comment more characteristic of the target object's preference, which can significantly improve the probability of the recommended comment being adopted and improve the accuracy of the generated second recommended comment. Displaying the second recommended comment in the predetermined order of the second recommended comment and the first recommended comment can prompt the target object to pay attention to the comments of interest, which helps to simplify the process of the target object posting comments and improve the user experience.

[0127] In the disclosed embodiment, the recommended comments displayed to the target object may be displayed in any of the following ways:

[0128] Method 1: In the browsing interface of the target resource, the first recommended comment is displayed in a floating window.

[0129] like Figure 8 As shown, the floating window will display the first recommended comment in the form of a small window on the browsing interface of the target resource, and the target object can choose to update the comment or send the comment by clicking the operation option therein.

[0130] When implemented, the floating window can be automatically displayed when the target object meets the preset conditions, for example, the browsing time of the target object in the current resource interface reaches the preset threshold; or the user manually triggers the display. The opening and closing of the floating window can support the on-demand control of the target object.

[0131] In the disclosed embodiment, recommended comments are displayed to the target object in the form of a floating window. The recommended comments in the floating window can serve as a topic guide to help the target object quickly find discussion points, so that the target object can pay attention to the comments and then make comments based on the recommended comments, thereby simplifying the operation process of the target object. In addition, the design of the floating window does not occupy the main content area, but exists as an auxiliary information window, which can also reduce interference with the content to be displayed by the target resource.

[0132] Method 2: Display the first recommended comment in the comment input box in the browsing interface of the target resource.

[0133] like Fig. 9 As shown, when the target object clicks the comment input box, the first recommended comment will be displayed in the input box. The target object can directly edit or delete it and then enter his own comment.

[0134] In the disclosed embodiment, at least one recommended comment is displayed in the comment input box in the browsing interface of the target resource, which can provide reference information to the target object, help the target object edit or post comments in the comment box, reduce the pressure of conceiving comments from scratch, and improve the user experience of posting comments.

[0135] Method three: when there are at least two first recommended comments, the at least two first recommended comments are displayed in a carousel.

[0136] like Fig.10 As shown, a small carousel area related to the comment input box is provided at the bottom of the target resource interface. By setting a corresponding time interval, when there are multiple first recommended comments, the first recommended comments generated are automatically carouseled (not shown in the figure). It is understandable that the carousel display can be in the aforementioned floating window or in the comment box. The order of the carousel can be in the order of recommendation.

[0137] In the disclosed embodiment, by displaying at least one recommended comment in a carousel, the target object can be helped to see viewpoints and ideas from different angles, and reference information from more angles can be provided to the target object, thereby stimulating the target object's interest in commenting.

[0138] Method 4: when there are at least two first recommended comments, in response to a comment switching operation of the target object, the at least two first recommended comments are displayed in sequence.

[0139] Continue with Figure 8 For example, after the target object clicks the "Change" switch operation, the recommended comments will be displayed in sequence, so that the recommended comments of interest can be searched based on the target object's wishes, thereby helping the target object to post comments and improving the user experience.

[0140] It is understandable that, when the process of displaying the recommended comments to the target object, in the case where the second recommended comment generated by the second largest model is obtained, the second recommended comment is displayed preferentially.

[0141] That is, as explained above, at least one second recommended comment generated online is inserted into the sequence of at least one first recommended comment generated by the first large model to be recommended to the target object.

[0142] In the disclosed embodiment, the second largest model generates the second recommended comments online, so the generated second recommended comments can better reflect the needs of the target object. The second recommended comments are displayed first, so that the target object can see the information most relevant to the current resource at the first time, thereby improving the efficiency of the target object in posting comments and improving the user experience.

[0143] During implementation, the selected recommended comment can be edited by responding to the target object's editing operation on the selected recommended comment to obtain the editing result.

[0144] It is understandable that each target object may have different ways and focuses of expression. Therefore, for the generated recommendation comments, the disclosed embodiment can support the target object to perform secondary editing operations according to his own ideas and preferences, so that it is more in line with his personal language habits and the specific information he wants to convey, thereby better meeting his personalized expression needs. This will enhance the sense of participation of the target object, thereby improving the satisfaction of the target object.

[0145] In some embodiments, the currently displayed recommended comments can also be sent by responding to the target object's request to send the currently displayed recommended comments. By sending the currently displayed recommended comments, the target object's needs for instant sharing can be met, allowing the target object to quickly pass on the content that he or she approves to others, enhancing the target object's sense of participation and satisfaction on the platform. In addition, the cumbersomeness of the target object's operation can be reduced, without the need for additional input or editing, saving time and energy, and making the comment on the target resource more convenient and efficient.

[0146] In summary, the comment processing method proposed in the present disclosure can be understood as generating recommended comments through a comment generation mechanism that combines online and offline streams, and then recommending them to the target object.

[0147] Specifically, Fig.11 As shown:

[0148] In S1101, when a target resource is newly released, an offline multi-style comment candidate generation mechanism is triggered so as to input the target resource features and multiple comment style templates into the first large model.

[0149] In S1102, the first large model generates candidate comments with various styles to enrich the comment candidate library. Then in S1103, the recall sorting model is used to screen and sort the candidate comments in the comment candidate library based on the first object feature of the target object, and the candidate comments that are more in line with the preferences of the target object are selected to improve the quality of the recommended comments. Subsequently, in S1104, the recommended comments adopted by the target object can be deleted based on the user feedback optimization system, and the generation mechanism can be triggered to supplement the comments in the comment candidate library to ensure the diversity of the candidate comments in the candidate comment library. At the same time, user feedback from the target object can also help further tune the recall sorting model to optimize the sorting effect to form a virtuous circle.

[0150] The online stream is built for high-rated users, aiming to provide more personalized comment generation capabilities. It can collect the target object's recent comment history at a daily granularity to provide data support for the second largest model when generating comments. This allows the second largest model to follow the target object's comment style and interests in real time. Fig.11As shown, in S1105, when the target object's browsing resource residence time reaches a threshold, the target object characteristics, resource characteristics, and the target object's comment history are obtained in real time as input to the second largest model. Then, in S1106, a recommended comment with personal characteristics is generated for the target object based on the second largest model. The client will give priority to displaying these personalized comments in the comment carousel.

[0151] Through the solution provided by the present disclosure, comment copy can be automatically generated, which greatly reduces the workload of the operation team and reduces labor costs. At the same time, highly personalized pre-made comment copy is intelligently generated based on the resource content and the historical comment habits of the target object. This customized experience can better touch the target object and stimulate the desire to comment, thereby significantly improving the quantity and quality of comments; on the other hand, the large model can quickly capture hot events and popular online terms, and the generated recommended comments are more likely to cause interaction between objects in the comment area, thereby driving the interaction and duration of the comment area. In addition, by intelligently generating diverse comment copy, it not only encourages the target object to comment, but also promotes the richness and diversity of the comment content, which is conducive to building a more active and healthy content ecology.

[0152] Based on the same technical concept, the embodiment of the present disclosure also provides a comment processing device 1200, such as Fig.12 As shown, including:

[0153] A generating module 1201, for generating at least two candidate reviews of a target resource using a first large model;

[0154] A screening module 1202, configured to screen a first recommended review corresponding to a target object from at least two candidate reviews;

[0155] The recommendation module 1203 is used to recommend a first recommended comment to the target object in response to the comment acquisition request.

[0156] In some embodiments, the generating module comprises:

[0157] A first acquisition unit, used to acquire resource characteristics of a target resource;

[0158] A generating unit is used to process resource features based on the first large model and at least two comment style templates to generate at least two candidate comments with at least two comment styles.

[0159] In some embodiments, the screening module comprises:

[0160] A ranking unit, configured to rank at least two candidate comments based on a first object feature of a target object and a resource feature of a target resource;

[0161] The screening unit is used to screen the first recommended comment based on the ranking of at least two candidate comments.

[0162] In some embodiments, the sorting unit is specifically configured to:

[0163] Obtaining semantic features of at least two candidate comments;

[0164] Inputting the first object feature, the resource feature, and the semantic feature of each candidate comment into the recall ranking model to obtain the target object's adoption probability of each candidate comment;

[0165] At least two candidate reviews are ranked based on the adoption probability.

[0166] In some embodiments, the generating module is specifically used to:

[0167] In case of publishing the target resource, at least two candidate comments are generated using the first model.

[0168] In some embodiments, the apparatus further comprises:

[0169] When the target resource is released, obtain the popularity change of the target resource;

[0170] Generate modules, including:

[0171] When the value of the heat change reaches a preset threshold, the first large model is used to generate at least two candidate comments.

[0172] In some embodiments, the apparatus further comprises:

[0173] a supplementing module, for, when the target resource is displayed and the target object belongs to a preset category, using a second large model to process a second object feature of the target object and a resource feature of the target resource to generate a second recommended comment corresponding to the target object;

[0174] A processing module, used for recommending a second recommended comment to a target object;

[0175] The first recommended comment and the second recommended comment are recommended to the target object in a predetermined order.

[0176] In some embodiments, the second object feature includes a comment record of the target object.

[0177] In some embodiments, the supplement module is specifically used to:

[0178] When it is determined that the display time of the target resource by the terminal device of the target object reaches the specified time and the target object belongs to a preset category, the second largest model is used to process the second object feature and the resource feature.

[0179] Based on the same technical concept, the embodiment of the present disclosure also provides a comment processing device 1300, such as Fig.13 As shown, including:

[0180] The acquisition module 1301 is used to acquire a first recommended comment of a target resource; the first recommended comment is generated based on the first large model and is selected for the target object from at least two candidate comments generated by the first large model;

[0181] The display module 1302 is used to display the first recommended comment when the terminal device of the target object displays the target resource.

[0182] In some embodiments, the apparatus further comprises:

[0183] A second acquisition unit is used to acquire a second recommended comment generated by a second large model in combination with a second object feature of the target object;

[0184] The display unit is used to display the second recommended comments according to a predetermined order of the second recommended comments and the first recommended comments.

[0185] In some embodiments, the first recommended review is displayed based on at least one of the following methods:

[0186] In the browsing interface of the target resource, the first recommended comment is displayed in a floating window;

[0187] Displaying the first recommended comment in the comment input box in the browsing interface of the target resource;

[0188] In the case of multiple first recommended comments, the multiple first recommended comments are displayed in a carousel.

[0189] In the case where there are a plurality of first recommended comments, in response to the target object's comment switching operation, the plurality of first recommended comments are displayed in sequence.

[0190] In some embodiments, the display unit is further used to:

[0191] When the second recommended comment generated by the second largest model is obtained, the second recommended comment is displayed first.

[0192] For the description of specific functions and examples of each module and submodule of the device in the embodiment of the present disclosure, reference can be made to the relevant description of the corresponding steps in the above method embodiment, which will not be repeated here.

[0193] In the technical solution disclosed herein, the acquisition, storage and application 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.

[0194] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0195] Fig.14 A schematic block diagram of an example electronic device 1400 that can be used to implement an embodiment of the present disclosure is shown. 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 can also represent various forms of mobile devices, such as personal digital assistants, 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 required herein.

[0196] like Fig.14 As shown, the device 1400 includes a computing unit 1401, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1402 or a computer program loaded from a storage unit 1408 into a random access memory (RAM) 1403. In the RAM 1403, various programs and data required for the operation of the device 1400 can also be stored. The computing unit 1401, the ROM 1402, and the RAM 1403 are connected to each other via a bus 1404. An input / output (I / O) interface 1405 is also connected to the bus 1404.

[0197] A number of components in the device 1400 are connected to the I / O interface 1405, including: an input unit 1406, such as a keyboard, a mouse, etc.; an output unit 1407, such as various types of displays, speakers, etc.; a storage unit 1408, such as a disk, an optical disk, etc.; and a communication unit 1409, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 1409 allows the device 1400 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0198] The computing unit 1401 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1401 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 1401 performs the various methods and processes described above, such as the comment processing method. For example, in some embodiments, the comment processing method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 1408. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 1400 via the ROM 1402 and / or the communication unit 1409. When the computer program is loaded into the RAM 1403 and executed by the computing unit 1401, one or more steps of the comment processing method described above may be performed. Alternatively, in other embodiments, the computing unit 1401 may be configured to perform the comment processing method in any other appropriate manner (e.g., by means of firmware).

[0199] 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), load 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.

[0200] 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.

[0201] 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 equipment, 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.

[0202] 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).

[0203] 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.

[0204] A computer system may include a client and a server. The client and the server 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 with each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0205] 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.

[0206] 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 principles of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A comment processing method, comprising: Using the first model to generate at least two candidate reviews of the target resource; Filtering a first recommended comment corresponding to the target object from the at least two candidate comments; In response to the comment acquisition request, the first recommended comment is recommended to the target object.

2. The method according to claim 1, wherein: The step of using the first large model to generate at least two candidate reviews for the target resource includes: Acquire resource characteristics of the target resource; The resource features are processed based on the first large model and at least two comment style templates to generate the at least two candidate comments with at least two comment styles.

3. The method according to claim 1, wherein: The step of selecting a first recommended comment corresponding to the target object from the at least two candidate comments includes: sorting the at least two candidate comments based on the first object feature of the target object and the resource feature of the target resource; The first recommended comment is filtered based on the ranking of the at least two candidate comments.

4. The method according to claim 3, wherein: The step of ranking the at least two candidate comments based on the first object feature of the target object and the resource feature of the target resource includes: Obtaining semantic features of each of the at least two candidate comments; Inputting the first object feature, the resource feature, and the semantic feature of each candidate comment into a recall ranking model to obtain the probability of the target object adopting each candidate comment; The at least two candidate comments are ranked based on the adoption probability.

5. The method according to claim 1, wherein: The step of using the first large model to generate at least two candidate reviews for the target resource includes: When the target resource is released, the at least two candidate comments are generated using the first large model.

6. The method according to claim 1, further comprising: When the target resource is released, obtaining a popularity change of the target resource; The step of using the first large model to generate at least two candidate reviews for the target resource includes: When the value of the heat change reaches a preset threshold, the first large model is used to generate the at least two candidate comments.

7. The method according to any one of claims 1 to 6, further comprising: When the target resource is displayed and the target object belongs to a preset category, a second large model is used to process a second object feature of the target object and a resource feature of the target resource to generate a second recommended comment corresponding to the target object; recommending the second recommended comment to the target object; The first recommended comment and the second recommended comment are recommended to the target object in a predetermined order. The method according to claim 7 , wherein the second object feature comprises a comment record of the target object.

9. The method according to claim 7, wherein: When the target resource is displayed and the target object belongs to a preset category, using a second large model to process a second object feature of the target object and a resource feature of the target resource to generate a second recommended comment corresponding to the target object includes: When it is determined that the display time of the target resource by the terminal device of the target object reaches a specified time and the target object belongs to a preset category, the second large model is used to process the second object features and the resource features to generate a second recommended comment corresponding to the target object.

10. A method for processing comments, comprising: Get the first recommended review of the target resource; The first recommended review is generated based on the first large model and is screened for the target object from at least two candidate reviews generated by the first large model; When the target object's terminal device displays the target resource, the first recommended comment is displayed.

11. The method according to claim 10, further comprising: Obtain a second recommended comment generated by the second largest model in combination with a second object feature of the target object; The second recommended comment is displayed in a predetermined order of the second recommended comment and the first recommended comment.

12. The method according to claim 10, wherein: The first recommended review is displayed based on at least one of the following methods: In the browsing interface of the target resource, displaying the first recommended comment in a floating window; Displaying the first recommended comment in a comment input box in a browsing interface of the target resource; In the case where there are at least two first recommended comments, displaying the at least two first recommended comments in a carousel; In the case where there are at least two first recommended comments, in response to a comment switching operation of the target object, the at least two first recommended comments are displayed in sequence.

13. The method according to claim 11, wherein: The displaying of the second recommended comments in a predetermined order of the second recommended comments and the first recommended comments includes: When the second recommended comment generated by the second largest model is obtained, the second recommended comment is displayed preferentially.

14. A comment processing device, comprising: A generating module, configured to generate at least two candidate reviews of a target resource using the first large model; A screening module, used for screening a first recommended review corresponding to the target object from the at least two candidate reviews; The recommendation module is used to recommend the first recommended comment to the target object in response to the comment acquisition request.

15. The device according to claim 14, wherein: The generating module comprises: A first acquisition unit, configured to acquire resource characteristics of the target resource; A generating unit is used to process the resource feature based on the first large model and at least two comment style templates to generate the at least two candidate comments with at least two comment styles.

16. The device according to claim 14, wherein: The screening module comprises: A sorting unit, configured to sort the at least two candidate comments based on the first object feature of the target object and the resource feature of the target resource; A screening unit is used to screen the first recommended comment based on the ranking of the at least two candidate comments.

17. The device according to claim 16, wherein: The sorting unit is specifically used for: Obtaining semantic features of each of the at least two candidate comments; Inputting the first object feature, the resource feature, and the semantic feature of each candidate comment into a recall ranking model to obtain the probability of the target object adopting each candidate comment; The at least two candidate comments are ranked based on the adoption probability.

18. The device according to claim 14, wherein: The generation module is specifically used for: When the target resource is released, the at least two candidate comments are generated using the first large model.

19. The apparatus according to claim 14, further comprising: When the target resource is released, obtaining a popularity change of the target resource; The generation module is specifically used for: When the value of the heat change reaches a preset threshold, the first large model is used to generate the at least two candidate comments.

20. The device according to any one of claims 14 to 19, further comprising: a supplementing module, configured to, when the target resource is displayed and the target object belongs to a preset category, use a second large model to process a second object feature of the target object and a resource feature of the target resource to generate a second recommended comment corresponding to the target object; A processing module, configured to recommend the second recommended comment to the target object; The first recommended comment and the second recommended comment are recommended to the target object in a predetermined order.

21. The apparatus according to claim 20, wherein the second object feature comprises a comment record of the target object.

22. The device according to claim 20, wherein: The supplementary module is specifically used for: When it is determined that the display time of the target resource by the terminal device of the target object reaches a specified time and the target object belongs to a preset category, the second large model is used to process the second object features and the resource features to generate a second recommended comment corresponding to the target object.

23. A comment processing device, comprising: An acquisition module, used to acquire the first recommended comment of the target resource; The first recommended review is generated based on the first large model and is screened for the target object from at least two candidate reviews generated by the first large model; A display module is used to display the first recommended comment when the terminal device of the target object displays the target resource.

24. The apparatus according to claim 23, further comprising: A second acquisition unit is used to acquire a second recommended comment generated by a second large model in combination with a second object feature of the target object; A display unit is used to display the second recommended comments in a predetermined order of the second recommended comments and the first recommended comments.

25. The device according to claim 23, wherein: The display module is specifically configured to display the first recommended comment based on at least one of the following methods: In the browsing interface of the target resource, displaying the first recommended comment in a floating window; Displaying the first recommended comment in a comment input box in a browsing interface of the target resource; In the case of multiple first recommended comments, displaying the multiple first recommended comments in a carousel; In the case where there are a plurality of first recommended comments, in response to a switch comment operation of the target object, the plurality of first recommended comments are displayed in sequence.

26. The device according to claim 24, wherein: The display unit is specifically used for: When the second recommended comment generated by the second largest model is obtained, the second recommended comment is displayed preferentially.

27. 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 13.

28. 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 1-13.

29. 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 13.

Images