Recommendation information generation method and device and related product

Abstract

Embodiments of the invention provide a recommendation information generation method and apparatus, and a related product. The method comprises the steps of obtaining first dialogue information of first video data and second dialogue information of second video data; the first video data and the second video data are located in the same video data set; the playing sequence of the first video data in the video data set is before the playing sequence of the second video data; generating a content summary of the first video data based on the first dialogue information of the first video data through a first generative model, and taking the content summary as a previous summary of the second video data; and through a second generative model, generating target recommendation information corresponding to the second video data based on the preceding summary of the second video data and the second dialogue information of the second video data.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technologies, and in particular, to a method, an apparatus, and related products for generating recommendation information. Background Art

[0002] In related technologies, a variety of video data is provided for users to select and watch. However, with the increase in the number of video data, it is difficult for users to determine the video data of interest from a large amount of video data in a short time. Therefore, it is necessary to provide recommendation information for video data to facilitate users to quickly select the video data of interest. Currently, recommendation information is mainly written manually for each video data, which has the disadvantage of low efficiency in generating recommendation information. Summary of the Invention

[0003] Embodiments of the present disclosure provide a method, an apparatus, and related products for generating recommendation information, which can automatically generate recommendation information for video data by using a generative model, and improve the efficiency of generating recommendation information.

[0004] In a first aspect, an embodiment of the present disclosure provides a method for generating recommendation information, including: Obtaining first dialogue information of a first video data and second dialogue information of a second video data; the first video data and the second video data are in the same video data set; the playing order of the first video data in the video data set is before the playing order of the second video data; Generating, by a first generative model, a content summary of the first video data based on the first dialogue information of the first video data, and using the content summary as a preview of the second video data; Generating, by a second generative model, a target recommendation information corresponding to the second video data based on the preview of the second video data and the second dialogue information of the second video data.

[0005] In a second aspect, an embodiment of the present disclosure provides a device for generating recommendation information, including: An information acquisition unit, configured to obtain first dialogue information of a first video data and second dialogue information of a second video data; the first video data and the second video data are in the same video data set; the playing order of the first video data in the video data set is before the playing order of the second video data; A content summary unit, configured to generate, by a first generative model, a content summary of the first video data based on the first dialogue information of the first video data, and use the content summary as a preview of the second video data; An information generation unit, configured to generate target recommendation information corresponding to the second video data based on the summary of the second video data and the second dialogue information of the second video data through a second generative model.

[0006] In a third aspect, an embodiment of the present disclosure provides an electronic device, including: a processor; and a memory configured to store computer-executable instructions, where the computer-executable instructions, when executed, cause the processor to implement the method described in the first aspect above.

[0007] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium for storing computer-executable instructions, where the computer-executable instructions, when executed by a processor, implement the method described in the first aspect above.

[0008] In a fifth aspect, an embodiment of the present disclosure provides a computer program product, including a computer program, where the computer program, when executed by a processor, implements the method described in the first aspect above.

[0009] In this embodiment, first, the first dialogue information of the first video data and the second dialogue information of the second video data are obtained. The first video data and the second video data are in the same video data set, and the playback order of the first video data in the video data set is before that of the second video data. Then, through the first generative model, based on the first dialogue information of the first video data, a content summary of the first video data is generated, and the content summary is used as the summary of the second video data. Finally, through the second generative model, based on the summary of the second video data and the second dialogue information of the second video data, target recommendation information corresponding to the second video data is generated. It can be seen that through this embodiment, the summary of the second video data and the second dialogue information of the second video data can be automatically generated through the generative model, and the target recommendation information corresponding to the second video data can be automatically generated based on the summary of the second video data and the second dialogue information of the second video data, achieving the effect of automatically generating recommendation information for video data by using the generative model and improving the generation efficiency of the recommendation information. Description of the Drawings

[0010] In order to more clearly illustrate the technical solutions in one or more embodiments of the present disclosure or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are only some embodiments recorded in the present disclosure. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts; Figure 1Flow chart of the recommendation information generation method provided by an embodiment of the present disclosure; Figure 2 Flow chart of the recommendation information generation method provided by another embodiment of the present disclosure; Figure 3 Structural diagram of the recommendation information generation device provided by an embodiment of the present disclosure; Figure 4 Structural diagram of the electronic device provided by an embodiment of the present disclosure. Detailed implementation manners

[0011] In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of the present disclosure, the following will clearly and completely describe the technical solutions in one or more embodiments of the present disclosure with reference to the accompanying drawings in one or more embodiments of the present disclosure. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all the embodiments. Based on one or more embodiments of the present disclosure, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present disclosure.

[0012] It can be understood that before using the technical solutions disclosed in the embodiments of the present disclosure, the relevant parties shall be informed of the type, scope of use, usage scenarios, etc. of the information involved in the present disclosure in an appropriate manner and obtain the authorization of the relevant parties in accordance with relevant laws and regulations.

[0013] For example, when responding to receiving an active request from a user, a prompt message is sent to the user to clearly prompt the user that the operation requested by the user will require obtaining and using the user's personal information. Thus, the user can autonomously choose whether to provide personal information to software or hardware such as an electronic device, an application program, a server, or a storage medium that executes the operation of the technical solution of the present disclosure according to the prompt message.

[0014] As an optional but non-limiting implementation manner, the manner of sending a prompt message to the user in response to receiving an active request from the user may be, for example, in the form of a pop-up window, and the prompt message may be presented in text in the pop-up window. In addition, the pop-up window may also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0015] It can be understood that the above process of notifying and obtaining user authorization is only illustrative and does not limit the implementation manner of the present disclosure. Other manners that meet relevant laws and regulations can also be applied to the implementation manner of the present disclosure.

[0016] Embodiments of the present disclosure provide a method, apparatus, and related products for generating recommendation information, which can automatically generate recommendation information for video data by using a generative model, improving the generation efficiency of recommendation information. Among them, the recommendation information method can be applied to a server side, implemented by the server side, and the server side includes both a single-server form and a server cluster form.

[0017] Figure 1 FIG. is a schematic flow chart of a method for generating recommendation information provided by an embodiment of the present disclosure. As Figure 1 shown, the process includes: Step S102, obtaining first dialogue information of first video data and second dialogue information of second video data; the first video data and the second video data are in the same video data set; the playing order of the first video data in the video data set is before the playing order of the second video data; Step S104, through a first generative model, based on the first dialogue information of the first video data, generating a content summary of the first video data, and using the content summary as a preview of the second video data; Step S106, through a second generative model, based on the preview of the second video data and the second dialogue information of the second video data, generating target recommendation information corresponding to the second video data.

[0018] In this embodiment, first, obtain the first dialogue information of the first video data and the second dialogue information of the second video data. The first video data and the second video data are in the same video data set, and the playing order of the first video data in the video data set is before the playing order of the second video data. Then, through the first generative model, based on the first dialogue information of the first video data, generate a content summary of the first video data, and use the content summary as a preview of the second video data. Finally, through the second generative model, based on the preview of the second video data and the second dialogue information of the second video data, generate target recommendation information corresponding to the second video data. It can be seen that through this embodiment, it is possible to automatically generate a preview of the second video data and the second dialogue information of the second video data, and automatically generate target recommendation information corresponding to the second video data based on the preview of the second video data and the second dialogue information of the second video data, achieving the effect of automatically generating recommendation information for video data by using a generative model, and improving the generation efficiency of recommendation information.

[0019] In the above step S102, the first dialogue information of the first video data and the second dialogue information of the second video data are obtained. The first video data and the second video data are in the same video data set. In the video data set, the playing order of the first video data is before that of the second video data. Here, the video data set can be exemplified as a short drama, a TV series, etc. The first video data and the second video data are each episode of the short drama or TV series. Moreover, in the short drama or TV series, the playing order of the first video data is before that of the second video data.

[0020] In a specific example, the first video data and the second video data are different episodes of the same short drama. The second video data is an episode of the short drama other than the first episode, and the first video data includes the previous one or more episodes before the second video data, such as the first five episodes. For example, the first video data and the second video data are in the same short drama. The second video data is the 10th episode of the short drama, and the first video data includes the 5th to 9th episodes of the short drama.

[0021] In another specific example, the first video data and the second video data are different episodes of the same short drama. The second video data is an episode of the short drama other than the first episode, and the first video data includes all the episodes before the second video data. For example, the first video data and the second video data are in the same short drama. The second video data is the 10th episode of the short drama, and the first video data includes the 1st to 9th episodes of the short drama.

[0022] In the above step S102, the first dialogue information of the first video data and the second dialogue information of the second video data are also obtained respectively. The first dialogue information of the first video data refers to the dialogue information of the characters in the first video data. The second dialogue information of the second video data refers to the dialogue information of the characters in the second video data. For example, if the first video data and the second video data are in the same short drama, the first video data includes the 5th to 9th episodes of the short drama, and the second video data is the 10th episode of the short drama, then the first dialogue information includes the dialogue information of the characters in these 5 episodes from the 5th to 9th episodes, and the second dialogue information is the dialogue information of the characters in the 10th episode of the short drama.

[0023] In one embodiment, obtaining the first dialogue information of the first video data includes: Performing speech recognition on the first video data to obtain the first alternative dialogue information of the first video data, and recognizing the subtitles and speakers in the video frame of the first video data to obtain the second alternative dialogue information of the first video data; Using the second alternative dialogue information to adjust the first alternative dialogue information, and taking the adjusted first alternative dialogue information as the first dialogue information.

[0024] In this embodiment, first, the Automatic Speech Recognition (ASR) technology is used to perform speech recognition on the first video data to obtain the first alternative dialogue information of the first video data. The ASR technology can recognize the audio information in the first video data and generate corresponding text information, and this text information is the first alternative dialogue information. When the first video data includes multiple short dramas or multiple TV dramas, the format of the first alternative dialogue information can be exemplified as follows: Episode N: Timestamp 1 Speaker: Speech content; Timestamp 2 Speaker: Speech content; …… Timestamp T Speaker: Speech content; Episode M: Timestamp 1 Speaker: Speech content; Timestamp 2 Speaker: Speech content; …… Timestamp P Speaker: Speech content.

[0025] Among them, N and M are positive integers.

[0026] In one example, the first alternative dialogue information can be exemplified as follows: Episode 5: Timestamp 00:30 seconds Speaker 1: Hello, I'm Han Meimei; Timestamp 00:40 seconds Speaker 2: Hello, I'm Li Lei; …… Timestamp 1 minute 30 seconds Speaker 3: Let's have dinner; Episode 6: Timestamp 00:15 seconds Speaker 1: Li Lei, can I go to your home to play? Timestamp 00:40 seconds Speaker 2: Okay, come on; …… Timestamp 1 minute 50 seconds Speaker 3: I'm leaving, uncle.

[0027] In this embodiment, the subtitles and speakers in the video frame of the first video data are also recognized to obtain the second alternative dialogue information of the first video data. The subtitles and speakers in the video frame of the first video data can be recognized through a video understanding model to obtain the second alternative dialogue information of the first video data. The video understanding model can use OCR (Optical Character Recognition) technology to recognize the text in the video frame, and can also recognize the speakers in the video frame through video frame analysis technology. Since the subtitles of the first video data are displayed in the video frame, the subtitles of the first video data can be recognized through OCR technology.

[0028] Considering that when using OCR technology to recognize the text in the video frame, in addition to recognizing the subtitles, non-subtitle text such as house numbers and store names in the frame may also be recognized. In this embodiment, the video understanding model uses OCR technology to recognize the text in the video frame, obtains the subtitles and non-subtitle text recognized by the OCR technology, discards the non-subtitle text recognized by the OCR technology, and recognizes the speakers in the video frame through video frame analysis technology. For the subtitles recognized by the OCR technology, an association relationship among video timestamp - subtitle - speaker is established. The video understanding model also outputs corresponding text information to represent the recognized association relationships, and this text information is the second alternative dialogue information. The format and example of the second alternative dialogue information can refer to the format and example of the first alternative dialogue information, which will not be repeated here.

[0029] Next, considering that there are cases of incorrect text information generation and missed recognition of audio information when the ASR technology generates text information based on audio information, in this embodiment, the second alternative dialogue information is used to adjust the first alternative dialogue information, and the adjusted first alternative dialogue information is used as the first dialogue information.

[0030] It can be seen that through this embodiment, speech recognition can be performed on the first video data to obtain the first alternative dialogue information of the first video data, and content recognition can be performed on the video frame of the first video data to obtain the second alternative dialogue information of the first video data. The second alternative dialogue information is used to adjust the first alternative dialogue information, and the adjusted first alternative dialogue information is used as the first dialogue information. The content recognition result of the video frame is used to correct the speech recognition result of the video data, thereby improving the accuracy of the generated first dialogue information.

[0031] In one embodiment, using the second alternative dialogue information to adjust the first alternative dialogue information includes: Determine the target dialogue information that is in the second alternative dialogue information and not in the first alternative dialogue information, and supplement the target dialogue information in the first alternative dialogue information; Determine the conflicting information in the second alternative dialogue information that conflicts with the first alternative dialogue information, and adjust the first alternative dialogue information based on the conflicting information.

[0032] In this embodiment, first, compare the first alternative dialogue information with the second alternative dialogue information to determine the target dialogue information that is in the second alternative dialogue information and not in the first alternative dialogue information, and supplement the target dialogue information in the first alternative dialogue information. The target dialogue information can be a single dialogue information that the second alternative dialogue information has more than the first alternative dialogue information, or several dialogue texts that the second alternative dialogue information has more than the first alternative dialogue information.

[0033] Also, determine the conflicting information in the second alternative dialogue information that conflicts with the first alternative dialogue information. When dialogue information is recognized at the same timestamp in both the first alternative dialogue information and the second alternative dialogue information, but the dialogue information recognized at this timestamp in the first alternative dialogue information is different from the dialogue information recognized at this timestamp in the second alternative dialogue information, the dialogue information recognized at this timestamp in the second alternative dialogue information is the conflicting information. The conflicting information is in the second alternative dialogue information, and adjust the first alternative dialogue information based on the conflicting information.

[0034] In one embodiment, the first alternative dialogue information and the second alternative dialogue information can also be input into a generative model. The generative model can include a large language model LLM (Large Language Model), and use the generative model to implement the above process, and output the first dialogue information by the generative model. The generative model can be a model other than the first generative model and the second generative model.

[0035] Referring to the example of the previous first alternative dialogue information, assume the corresponding second alternative dialogue information is: Episode 5: Timestamp 0:00:40 Speaker 2: Hello, I'm Li Lei; …… Timestamp 1:30 Speaker 3: Everyone, come and have dinner quickly; Episode 6: Timestamp 0:00:15 Speaker 1: Li Lei, can I go to your home to play? Timestamp 0:00:40 Speaker 2: Okay, come on; Timestamp 0:00:42 Speaker 2: I'll wait for you; …… Timestamp 1 minute 50 seconds Speaker 1: I'm leaving, bye.

[0036] Comparing the first candidate dialogue information with the second candidate dialogue information, it can be seen that the second candidate dialogue information has one more dialogue information than the first candidate dialogue information: Episode 6 "Time stamp 0042 seconds Speaker 2: I'll wait for you"; this dialogue information is added to the first candidate dialogue information as the target dialogue information.

[0037] Comparing the first alternative dialogue information and the second alternative dialogue information, it can be seen that in the second alternative dialogue information, the dialogue information at 1 minute and 50 seconds in episode 6 is "Time stamp 1 minute and 50 seconds Speaker 1: I'm leaving, bye", however, in the first alternative dialogue information, the dialogue information at 1 minute and 50 seconds in episode 6 is "Time stamp 1 minute and 50 seconds Speaker 3: I'm leaving, uncle", therefore, the dialogue information "Time stamp 1 minute and 50 seconds Speaker 1: I'm leaving, bye" is contradictory information. Based on the contradictory information, the first alternative dialogue information is modified, and the dialogue information at 1 minute and 50 seconds in episode 6 in the first alternative dialogue information is modified to "Time stamp 1 minute and 50 seconds Speaker 1: I'm leaving, bye".

[0038] Comparing the first candidate dialogue information with the second candidate dialogue information, it can be seen that the second candidate dialogue information has several more dialogue texts than the first candidate dialogue information: the word "Quick" in the dialogue information of episode 5, 1 minute and 30 seconds "Time stamp 1 minute and 30 seconds Speaker 3: Let's all come to eat quickly", and the dialogue text "Quick" is added to the first candidate dialogue information as the target dialogue information. Of course, the dialogue information of episode 5, 1 minute and 30 seconds "Time stamp 1 minute and 30 seconds Speaker 3: Let's all come to eat quickly" in the second candidate dialogue information can also be treated as contradictory information, and the dialogue information of episode 5, 1 minute and 30 seconds in the first candidate dialogue information can be modified to the contradictory information.

[0039] Through the above process, the first dialogue information can be obtained as follows: Episode 5: Timestamp 0030 seconds Speaker 1: Hello, I am Han Meimei; Timestamp 0040 seconds Speaker 2: Hello, I am Li Lei; … Timestamp 1 minute 30 seconds Speaker 3: Let's have dinner soon; Episode 6: Timestamp 0015 seconds Speaker 1: Li Lei, can I come to your house to play? Timestamp 0040 seconds Speaker 2: Okay, come on; Timestamp 0042 seconds Speaker 2: I'll wait for you; … Timestamp: 1 minute 50 seconds, Speaker 1: I'm leaving. Bye.

[0040] As can be seen from the above example, compared with the first alternative dialogue information, the second alternative dialogue information lacks the dialogue information of "Timestamp: 00:30 seconds, Speaker 1: Hello, I'm Han Meimei" in Episode 5. The reason for the lack may be that there is dialogue information in the first video data but the corresponding subtitles are not displayed.

[0041] It can be seen that through this embodiment, by using the first alternative dialogue information as a reference and adjusting the first alternative dialogue information based on the second alternative dialogue information, the advantage of being able to recognize the speech content in the first video data by speech recognition is achieved, avoiding the missed recognition of the second alternative dialogue information caused by the non-display of subtitles. Moreover, using the second alternative dialogue information to supplement the missing dialogue information in the first alternative dialogue information and correct the misrecognized dialogue information in the first alternative dialogue information, such as dialogue text or speaker, can supplement and correct the missed recognition and misrecognition of speech recognition, and improve the accuracy of generating the first dialogue information.

[0042] In one embodiment, the process of obtaining the second dialogue information of the second video data can be the same as the process of obtaining the first dialogue information. For example, it includes: Performing speech recognition on the second video data to obtain the third alternative dialogue information of the second video data, and recognizing the subtitles and speakers in the video frame of the second video data to obtain the fourth alternative dialogue information of the second video data; adjusting the third alternative dialogue information by using the fourth alternative dialogue information, and using the adjusted third alternative dialogue information as the second dialogue information. Among them, adjusting the third alternative dialogue information by using the fourth alternative dialogue information and using the adjusted third alternative dialogue information as the second dialogue information includes: determining the target dialogue information that is in the fourth alternative dialogue information and not in the third alternative dialogue information, and supplementing the target dialogue information in the third alternative dialogue information; determining the contradictory information in the fourth alternative dialogue information that is contradictory to the third alternative dialogue information, and adjusting the third alternative dialogue information based on the contradictory information. The process of obtaining the second dialogue information of the second video data will not be further explained here and can refer to the process of obtaining the first dialogue information. In this embodiment, the first video data includes one or more episodes of short dramas or TV series, and the second video data can be one episode of a short drama or one episode of a TV series. Therefore, when obtaining the second dialogue information, it is equivalent to obtaining the dialogue information of one episode of a short drama or one episode of a TV series through the above process.

[0043] After obtaining the first conversation information and the second conversation information, in step S104 above, through the first generative model, based on the first conversation information of the first video data, generate a content summary of the first video data, and use the content summary as the preview of the second video data. In this embodiment, the content summary of the first video data can be obtained by summarizing the first conversation information of the first video data through the first generative model. Since the playback order of the first video data is before that of the second video data, the content summary of the second video data can be used as the preview of the second video data. The first generative model can include a large language model.

[0044] In one embodiment, generating a content summary of the first video data based on the first conversation information of the first video data through the first generative model includes: Obtain a summary generation prompt word corresponding to the content summary; the summary generation prompt word is used to represent the content summary requirements corresponding to the content summary and the content summary example corresponding to the content summary; Through the first generative model, summarize the information content of the first conversation information according to the content summary requirements and the content summary example to obtain the content summary of the first video data.

[0045] In this embodiment, obtain a summary generation prompt word corresponding to the content summary of the first video data, and the summary generation prompt word is used to represent the content summary requirements corresponding to the content summary of the first video data and the content summary example corresponding to the content summary of the first video data. In one example, the content summary requirements include: Please generate a content summary for each episode, with the number of words in each episode's content summary not exceeding 100 words; the content summary of each episode needs to briefly describe the content of each episode according to the conversation information, without completely covering all the content of this episode, but there should be no plot errors; do not add synonymous words such as "this episode" at the beginning of the content summary of each episode, and there should be no suspense at the end of the content summary of each episode, and do not add words such as "anticipating" etc. The content summary example can be exemplified as: In episode 1, classmate X studied hard actively in order to be admitted to a prestigious university. Unfortunately, he fainted during an evening self-study class and was sent to the hospital in time by his classmates and teachers; In episode 2, after being diagnosed by the doctor, classmate X suffered from acute leukemia and needed a large amount of medical expenses; In episode 3, through the efforts of the school, classmates and teachers donated money to classmate X one after another, and classmate X quickly recovered and returned to class; In episode 4, through unremitting efforts, classmate X was admitted to his ideal institution.

[0046] In this embodiment, input the first conversation information and the summary generation prompt word into the first generative model, and through the first generative model, summarize the information content of the first conversation information according to the content summary requirements and the content summary example to obtain the content summary of the first video data. When the first video data includes multiple short dramas or multiple episodes of a TV series, the content summary of the first video data includes the content of each short drama or TV series episode respectively.

[0047] It can be seen that through this embodiment, a summary generation prompt word corresponding to the content summary of the first video data can be obtained. Through the first generative model, according to the content summary requirements and content summary examples indicated by the summary generation prompt word, the first dialogue information is summarized to obtain the content summary of the first video data. By utilizing the excellent text processing ability and text analysis ability of the generative model, the accuracy and efficiency of generating the content summary of the first video data are improved.

[0048] After obtaining the content summary of the first video data and using the content summary as the preview of the second video data, in step S106 above, through the second generative model, based on the preview of the second video data and the second dialogue information of the second video data, the target recommendation information corresponding to the second video data is generated. The second generative model can be a model different from the first generative model. The second generative model can also include a large language model.

[0049] In this embodiment, the second video data can be any episode other than the first episode in a short drama or a TV series, and the first video data can include multiple previous episodes of the second video data or all episodes before the second video data. In one case, the target recommendation information is used to comment on the second video data. The target recommendation information can include bullet screen information or comment information of the second video data, and the target recommendation information can be published in the bullet screen area or comment area of the second video data. In another case, the target recommendation information is a post or article recommending the second video data, or a post or article recommending the short drama / TV series where the second video data is located. The post or article recommending the second video data or the short drama / TV series where the second video data is located can be published in a drama recommendation community. It can be seen that this embodiment can generate target recommendation information that is closely related to the single-episode plot and related to the previous plot based on the single-episode content and the preview of multiple previous episodes, making the target recommendation information more in line with the real feeling of watching the single episode and the previous multiple episodes, and improving the recommendation effect of the target recommendation information.

[0050] In this embodiment, the format of the preview can be set to be different from the format of the second dialogue information. For example, the format of the preview is a long text, and the format of the second dialogue information is in the form of a dialogue flow. And a prompt message is input to the second generative model to prompt it to focus on understanding the second dialogue information, so that the second generative model can distinguish the preview and the second dialogue information from the format, and generate target recommendation information that is closely related to the plot of the second video data and related to the plot of the first video data, reducing the possibility of the second generative model having hallucinations.

[0051] In one embodiment, the target recommendation information corresponding to the second video data is generated by the second generative model based on the context summary of the second video data and the second dialogue information of the second video data, including: Determine a recommended scene for the second video data; the recommended scene includes the first scene or the second scene; the first scene is used to comment on the characters and plot in the second video data; the second scene is used to recommend the second video data to the user; According to the recommendation scenario, a first generated prompt word corresponding to the target recommendation information is generated; the first generated prompt word is used to indicate the information content requirement corresponding to the target recommendation information; the information content requirement matches the recommendation scenario; Through the second generative model, according to the information content requirements, content expansion is performed based on the previous summary and the second dialogue information to generate target recommendation information.

[0052] In this embodiment, first, the recommended scene of the second video data is determined. The recommended scene may be the first scene or the second scene. The first scene is a scene for commenting on the characters and plots in the second video data. The first scene can be exemplified by the comment area of ​​the second video data, the bullet screen area of ​​the second video data, the discussion community corresponding to the short play / TV series where the second video data is located, etc. The user can post comment information on the characters and plots in the second video data in the comment area and bullet screen area of ​​the second video data. The user can also post comment information on the characters and plots in the second video data in the discussion community corresponding to the short play / TV series where the second video data is located.

[0053] The second scenario is a scenario for recommending the second video data or the video data (short play / TV series) where the second video data is located. In the second scenario, the user can publish recommendation information such as "This episode is really great" to recommend the second video data to other users, or publish recommendation information such as "This drama is great" to recommend the short play / TV series where the second video data is located to other users. The second scenario can be exemplified by a discussion community for recommending video data, such as a playlist, in which the user can recommend a short play / TV series that he or she is interested in or an episode thereof, such as recommending the short play / TV series where the second video data is located or recommending the second video data.

[0054] Next, according to the recommendation scene of the second video data, the information content requirement corresponding to the target recommendation information is determined, and the information content requirement matches the recommendation scene, and the information content requirement is used to constrain the content of the target recommendation information. In addition, a prompt word carrying the information content requirement is generated as the first generated prompt word corresponding to the target recommendation information. For example, when the recommendation scene of the second video data is the first scene, the information content requirement can be determined as: commenting on the characters and plots in the second video data, so that the generated target recommendation information can achieve the effect of commenting on the characters and plots in the second video data. For another example, when the recommendation scene of the second video data is the second scene, the information content requirement can be determined as: generating recommendation information for the entire second video data, and the recommendation information includes the plot introduction, highlight clip introduction, and after-viewing experience of the second video data and the short drama / TV series in which it is located, so as to achieve the effect of recommending the second video data to other users and the short drama / TV series in which the second video data is located. In the second scene, the generated target recommendation information is not limited to commenting on the plot and characters in the second video data, but achieves the effect of recommending the second video data to other users and the short drama / TV series in which the second video data is located.

[0055] Next, the previous situation summary, the second dialogue information and the first generated prompt word are input into the second generative model. Through the second generative model, the content is expanded based on the previous situation summary and the second dialogue information according to the information content requirements to generate target recommendation information. The target recommendation information meets the information content requirements. When the recommended scene includes the first scene, the target recommendation information can achieve the effect of commenting on the characters and plots in the second video data in the first scene. When the recommended scene includes the second scene, the target recommendation information can achieve the effect of recommending the second video data to other users in the second scene and recommending the short play / TV series in which the second video data is located.

[0056] It can be seen that through this embodiment, the first generation prompt word corresponding to the target recommendation information can be generated according to the recommendation scene of the second video data, and then the target recommendation information is generated, so that the target recommendation information matches the recommendation scene of the second video data, thereby improving the accuracy of the target recommendation information.

[0057] In one embodiment, generating a first generated prompt word corresponding to the target recommendation information according to the recommendation scenario includes: Acquire first sample recommendation information matching the recommendation scenario, and generate a first candidate generation prompt word corresponding to the target recommendation information according to the first sample recommendation information; the first candidate generation prompt word is used to prompt the second generative model to generate recommendation information with reference to the first sample recommendation information; Based on the information content requirement, the first candidate generated prompt word is tested and adjusted to obtain the first generated prompt word.

[0058] In this embodiment, first, obtain first sample recommendation information that matches the recommendation scenario. The first sample recommendation information can be high-quality recommendation information obtained manually in the recommendation scenario of the second video data.

[0059] Then, generate a first alternative generation prompt word corresponding to the target recommendation information according to the first sample recommendation information. The first alternative generation prompt word is used to prompt the second generative model to generate recommendation information by referring to the first sample recommendation information. For example, input the text "You are a large model application development engineer. Please generate a prompt word according to the following requirements to implement this requirement. The requirement is to generate recommendation information by referring to the first sample recommendation information, and the first sample recommendation information is XXXXX" into the second generative model. The second generative model can generate a prompt word for implementing the above requirements, and this prompt word is the first alternative generation prompt word. The first alternative generation prompt word can also be generated by the first generative model or other generative models.

[0060] Since the first alternative generation prompt word is a prompt word that can enable the second generative model to generate recommendation information by referring to the first sample recommendation information, but considering that the recommendation information generated according to the first alternative generation prompt word may not meet the above information content requirements, finally, based on the information content requirements, test and adjust the first alternative generation prompt word to obtain the first generation prompt word. Since the first generation prompt word is adjusted based on the information content requirements, the first generation prompt word can accurately represent the information content requirements, thereby improving the accuracy of the target recommendation information.

[0061] It can be seen that through this embodiment, obtain the first sample recommendation information that matches the recommendation scenario, generate the first alternative generation prompt word corresponding to the target recommendation information according to the first sample recommendation information. The first alternative generation prompt word is used to prompt the second generative model to generate recommendation information by referring to the first sample recommendation information. Based on the information content requirements, test and adjust the first alternative generation prompt word to obtain the first generation prompt word, so that the first generation prompt word can accurately represent the information content requirements, thereby improving the accuracy of the target recommendation information.

[0062] In one embodiment, based on the information content requirements, testing and adjusting the first alternative generation prompt word to obtain the first generation prompt word includes: Obtain the first test dialogue information and the first test context. Through the second generative model, generate the first test recommendation information according to the first alternative generation prompt word, the first test dialogue information, and the first test context; Based on the information content requirements, perform positive marking and negative marking on the first test recommendation information to obtain the marked first test recommendation information; Adjust the first alternative generation prompt according to the first test recommendation information after marking to generate the first generation prompt.

[0063] In this embodiment, first, obtain the first test dialogue information and the first test summary. The first test dialogue information and the first test summary can be the dialogue information written manually in advance or the dialogue information and the summary of the video data obtained in advance. Then, input the first alternative generation prompt, the first test dialogue information, and the first test summary into the second generative model. Through the second generative model, according to the input data, generate recommendation information, and the generated recommendation information is the first test recommendation information.

[0064] Then, based on the information content requirements, perform positive marking and negative marking on the first test recommendation information to obtain the first test recommendation information after marking. The first test recommendation information that meets the information content requirements can be positively marked, such as tagging "1", and the first test recommendation information that does not meet the information content requirements can be negatively marked, such as tagging "0", to obtain the first test recommendation information after marking.

[0065] Finally, input the first test recommendation information after marking and the first alternative generation prompt into the prompt adjustment platform. Through the prompt adjustment platform, adjust the first alternative generation prompt according to the first test recommendation information after marking to generate the first generation prompt. The prompt adjustment platform can determine the first test recommendation information that meets the information content requirements and the first test recommendation information that does not meet the information content requirements according to the positive marking and negative marking, and adjust the first alternative generation prompt in the direction of generating the recommendation information that meets the information content requirements, so as to obtain the first generation prompt. The first generation prompt may or may not carry an example of the target recommendation information.

[0066] In one example, the first test recommendation information can be generated using the first alternative generation prompt, and positive marking and negative marking are performed on the generated first test recommendation information based on the information content requirements. Then, through the prompt adjustment platform, the first alternative generation prompt is adjusted according to the first test recommendation information after marking to obtain the first alternative generation prompt after one adjustment. Then, use the first alternative generation prompt after one adjustment to generate the first test recommendation information again, and perform positive marking and negative marking on the first test recommendation information generated again. If the quantity ratio of the positively marked first test recommendation information is greater than the ratio threshold, then the first alternative generation prompt after one adjustment is used as the first generation prompt. If the quantity ratio of the positively marked first test recommendation information is not greater than the ratio threshold, repeat the above process until the quantity ratio of the positively marked first test recommendation information is greater than the ratio threshold to obtain the first generation prompt.

[0067] In one example, example recommendation information that meets the information content requirements can also be supplemented in the first alternative generation prompt word generated for the first time or the first alternative generation prompt word after a certain adjustment. The first test recommendation information is generated using the first alternative generation prompt word carrying the example recommendation information, and it is detected whether the quantity ratio of the first test recommendation information with positive markings in the generated first test recommendation information is greater than the ratio threshold, that is, it is detected whether the first alternative generation prompt word carrying the example recommendation information can be used as the first generation prompt word. By adding example recommendation information that meets the information content requirements to the first alternative generation prompt word, the effect of quickly adjusting the first alternative generation prompt word is achieved.

[0068] It can be seen that through this embodiment, the first generation prompt word can be accurately generated based on the first alternative generation prompt word and the information content requirements by testing and adjusting the first alternative generation prompt word.

[0069] In one embodiment, through the second generative model, content expansion is performed based on the summary and the second dialogue information according to the information content requirements to generate the target recommendation information, including: Through the second generative model, content extraction is performed on the summary according to the information content requirements to obtain the first extraction result, and content extraction is performed on the second dialogue information according to the information content requirements to obtain the second extraction result; Through the second generative model, content expansion is performed based on the first extraction result and the second extraction result to generate the target recommendation information.

[0070] In one case, if the information content requirement is to comment on the characters and plot in the second video data, so that the generated target recommendation information can achieve the effect of commenting on the characters and plot in the second video data, then the content related to the characters and plot is extracted from the summary as the first extraction result, and the content related to the characters and plot is extracted from the second dialogue information as the second extraction result. Then, content expansion is performed based on the first extraction result and the second extraction result to generate the target recommendation information. It can be understood that the first extraction result and the second extraction result may only include content related to the plot or only include content related to the characters. The target recommendation information can comment on either the characters or the plot in the second video data. Since there is an association relationship between the characters and plot in the second video data and the characters and plot in the previously played video data, generating the target recommendation information based on the summary and the second dialogue information can achieve the effect of accurately commenting on the characters and plot in the second video data. In one example, the target recommendation information in the first scenario can be exemplified as "The male lead's look in this episode is so handsome", "This episode's plot is so mind-bending, who can explain it to me", etc.

[0071] In another case, if the information content requirement is: generate recommendation information for the whole of the second video data, the recommendation information includes the plot introduction, highlight clip introduction, and after-viewing experience of the second video data and the short play / TV series in which it is located, so as to achieve the effect of recommending the second video data and the short play / TV series in which the second video data is located to other users, then the content of the plot, the content of the characters, the content of the highlight clips, the content of the causal relationship between the plots, etc. are extracted from the synopsis as the first extraction result, and the content of the plot, the content of the characters, the content of the highlight clips, the content of the causal relationship between the plots, etc. are extracted from the second dialogue information as the second extraction result. Then, the content is expanded according to the first extraction result and the second extraction result to generate the target recommendation information. The highlight clip refers to the wonderful clip in the video data that is liked by the user or the important clip that promotes the plot. The target recommendation information can include the plot introduction, highlight clip introduction, and after-viewing experience of the second video data and the short play / TV series in which it is located, so as to recommend the second video data and the short play / TV series in which the second video data is located to other users. By generating target recommendation information based on the previous synopsis and the second dialogue information, it is possible to generate accurate plot introductions, highlight clip introductions and after-viewing impressions in combination with the plots of each episode, so as to achieve the effect of accurately recommending the second video data and the short drama / TV series in which it is located. In an example, the target recommendation information in the second scenario can be exemplified as "This episode is simply the ceiling of the competition, it must be recommended", "This drama is worth collecting and watching repeatedly, it is too classic", etc. Of course, the second recommendation information can also be in the form of long text, such as posts or articles recommending TV series / short dramas, etc., which are not limited here.

[0072] It can be seen that through this embodiment, the second generative model can be used to extract content from the previous situation summary according to the information content requirements to obtain a first extraction result, and the second conversation information can be extracted according to the information content requirements to obtain a second extraction result. Through the second generative model, content expansion is performed based on the first extraction result and the second extraction result to generate target recommendation information, thereby accurately generating target recommendation information based on the recommendation scenario.

[0073] In one embodiment, when it is not necessary to distinguish the recommendation scene of the second video data, generating target recommendation information corresponding to the second video data based on the background summary of the second video data and the second dialogue information of the second video data by using the second generative model includes: Obtaining a second generated prompt word corresponding to the target recommended information; the second generated prompt word is used to indicate the recommended information requirements corresponding to the target recommended information; the recommended information requirements include information style requirements, information length requirements, information sentence requirements and information quantity requirements; Based on the second generative model, content expansion is performed according to the information style requirements, information length requirements, information sentence pattern requirements, and information quantity requirements, based on the previous summary and the second conversation information to generate target recommendation information.

[0074] The target recommendation information generated in this embodiment may include information for commenting on the second video data, may also include articles or posts for recommending the second video data, and may also include articles or posts for recommending the short drama / TV series where the second video data is located.

[0075] In this embodiment, first, obtain the second generation prompt word corresponding to the target recommendation information. The second generation prompt word is used to represent the recommendation information requirements corresponding to the target recommendation information. The recommendation information requirements include information style requirements, information length requirements, information sentence pattern requirements, and information quantity requirements.

[0076] Next, input the previous summary of the second video data, the second conversation information of the second video data, and the second generation prompt word into the second generative model. Through the second generative model, analyze the recommendation information requirements represented by the second generation prompt word to determine the information style requirements, information length requirements, information sentence pattern requirements, and information quantity requirements represented by the recommendation information requirements. Among them, the information style requirement is used to represent the requirement for the emotional style of the target recommendation information, such as neutral emotion, positive emotion, etc. Neutral emotion means that the emotional color of the sentence does not carry a strong subjective will, but only describes facts or asks questions. Positive emotion means that the emotional color of the sentence carries a strong subjective will of praise, either praising facts or reflecting the praise attitude through the form of asking questions. The information length requirement is used to represent the requirement for the text length of a single target recommendation information, such as a single target recommendation information includes 10 - 20 characters. The information sentence pattern requirement is used to represent the requirement for the sentence pattern of a single target recommendation information, such as declarative sentence, interrogative sentence, exclamatory sentence, etc. The information quantity requirement is used to represent the requirement for the number of target recommendation information generated, such as generating 10 target recommendation information.

[0077] In this embodiment, the recommendation information requirements are used to represent various information length requirements under each information style requirement, and represent various information sentence pattern requirements under each information length requirement, and represent the information quantity requirements under each information sentence pattern requirement. In one example, the recommendation information requirements can be referred to the following table. The information sentence pattern requirements are not limited to the declarative sentence and interrogative sentence in the following table, and can also be exclamatory sentences, etc.

[0078] Table 1

[0079] As shown in the above table, example information of the target recommendation information can also be input into the second generative model. This example information can be carried in the second generation prompt, so as to facilitate the generative model to understand the recommendation information requirements corresponding to the target recommendation information.

[0080] Next, through the second generative model, based on the determined information style requirements, information length requirements, information sentence pattern requirements, and information quantity requirements, content expansion is performed based on the previous summary and the second dialogue information to generate the target recommendation information. The target recommendation information should preferably meet the information style requirements, information length requirements, information sentence pattern requirements, and information quantity requirements. The second generative model can generate the target recommendation information by imitating the example of the target recommendation information. The target recommendation information is used to recommend the second video data, and the target recommendation information needs to describe the advantages of the second video data to achieve the recommendation effect. For example, describe that the plot is very exciting, or describe that the character modeling is very beautiful, etc. Thus, it can be seen that the content of the target recommendation information cannot be directly obtained from the second dialogue information and the previous summary, but needs to be expanded based on the second dialogue information and the previous summary. Therefore, it is necessary to use the second generative model to perform content expansion based on the previous summary and the second dialogue information to generate the target recommendation information for recommending the second video data.

[0081] In this embodiment, the name of the short drama or TV drama where the second video data is located, and the synopsis of the TV drama or short drama can also be input into the second generative model. Then, the second generative model performs content expansion based on the determined information style requirements, information length requirements, information sentence pattern requirements, and information quantity requirements, based on the previous summary, the second dialogue information, the drama name, and the drama synopsis, to generate the target recommendation information corresponding to the second video data.

[0082] It can be seen that through this embodiment, the information style requirements, information length requirements, information sentence pattern requirements, and information quantity requirements represented by the recommendation information requirements can be determined through the second generative model, and, through the second generative model, based on the information style requirements, information length requirements, information sentence pattern requirements, and information quantity requirements, content expansion is performed based on the previous summary and the second dialogue information to generate the target recommendation information, using the excellent text processing ability and text analysis ability of the generative model to improve the accuracy and efficiency of generating the target recommendation information, and accurately generate the target recommendation information.

[0083] As can be seen from the above process, it is necessary to input the above-mentioned summary, the above-mentioned second dialogue information, and the second generation prompt word corresponding to the target recommendation information into the second generative model, and generate the target recommendation information through the second generative model. The second generation prompt word is used to represent the recommendation information requirements corresponding to the target recommendation information, and the recommendation information requirements can refer to those shown in the above table. Therefore, whether the second generation prompt word can accurately represent the recommendation information requirements affects the accuracy of the generated target recommendation information. Therefore, it is necessary to generate accurate second generation prompt words to improve the accuracy of the target recommendation information.

[0084] In one embodiment, obtaining the second generation prompt word corresponding to the target recommendation information includes: Obtaining the second sample recommendation information, and generating the second alternative generation prompt word corresponding to the target recommendation information according to the second sample recommendation information; the second alternative generation prompt word is used to prompt the second generative model to generate recommendation information with reference to the second sample recommendation information; Based on the recommendation information requirements, testing and adjusting the second alternative generation prompt word to obtain the second generation prompt word.

[0085] In this embodiment, first, obtain the recommendation information specification and the second sample recommendation information corresponding to the recommendation information specification. The recommendation information specification can be a manually set specification for representing the high-quality recommendation information of video data, such as the standard for high-quality comments of video data. The second sample recommendation information corresponding to the recommendation information specification can be recommendation information collected or written manually, such as the comment information or bullet screen information of video data written manually. The second sample recommendation information meets the recommendation information specification.

[0086] Then, through the second generative model, generate the second alternative generation prompt word corresponding to the target recommendation information. The second alternative generation prompt word is used to prompt the second generative model to generate recommendation information in accordance with the recommendation information specification and with reference to the second sample recommendation information. The recommendation information generation requirement can be determined based on the recommendation information specification and the second sample recommendation information. The recommendation information generation requirement is used to represent the requirement for generating recommendation information in accordance with the recommendation information specification and with reference to the second sample recommendation information. Through the second generative model, generate the prompt word for realizing the recommendation information generation requirement, and use the generated prompt word as the second alternative generation prompt word corresponding to the target recommendation information.

[0087] For example, input the text "You are a large model application development engineer. Please generate a prompt according to the following requirements to meet the requirement of generating recommendation information in accordance with the recommendation information specification and referring to the second sample recommendation information. The recommendation information specification is XXXXXX, and the second sample recommendation information is XXXXXX" into the second generative model. Through the second generative model, a prompt for meeting the above requirements can be generated, and this prompt is the second alternative generation prompt corresponding to the target recommendation information. The second alternative generation prompt can also be generated through the first generative model or other generative models.

[0088] Since the second alternative generation prompt is a prompt that enables the second generative model to generate recommendation information in accordance with the recommendation information specification and referring to the second sample recommendation information, that is, the second alternative generation prompt can generate high-quality recommendation information for video data. However, the recommendation information generated according to the second alternative generation prompt may not meet the above-mentioned recommendation information requirements. Therefore, finally, based on the above-mentioned recommendation information requirements, the second alternative generation prompt is tested and adjusted to obtain the second generation prompt. Since the second generation prompt is adjusted based on the recommendation information requirements, the second generation prompt can accurately represent the recommendation information requirements, thereby improving the accuracy of the target recommendation information.

[0089] It can be seen that through this embodiment, based on the second alternative generation prompt that can generate high-quality recommendation information for video data, the second alternative generation prompt is tested and adjusted according to the recommendation information requirements to obtain the second generation prompt, so that the second generation prompt can accurately represent the recommendation information requirements, thereby improving the accuracy of the target recommendation information.

[0090] In one embodiment, based on the above-mentioned recommendation information requirements, testing and adjusting the second alternative generation prompt to obtain the second generation prompt includes: Obtain the second test dialogue information and the second test context. Through the second generative model, according to the second alternative generation prompt, the second test dialogue information, and the second test context, generate the second test recommendation information; Based on the recommendation information requirements, perform positive marking and negative marking on the second test recommendation information to obtain the marked second test recommendation information; Adjust the second alternative generation prompt according to the marked second test recommendation information to generate the second generation prompt.

[0091] In this embodiment, first, obtain the second test dialogue information and the second test synopsis. The second test dialogue information and the second test synopsis can be the dialogue information manually written in advance or the synopsis of the video data obtained in advance. Then, input the second alternative generation prompt, the second test dialogue information, and the second test synopsis into the second generative model. Through the second generative model, according to the input data, generate recommendation information, and the generated recommendation information is the second test recommendation information.

[0092] Then, based on the recommendation information requirements, perform positive marking and negative marking on the second test recommendation information to obtain the marked second test recommendation information. The second test recommendation information that meets the recommendation information requirements can be positively marked, such as tagging "1", and the second test recommendation information that does not meet the recommendation information requirements can be negatively marked, such as tagging "0", to obtain the marked second test recommendation information.

[0093] Finally, input the marked second test recommendation information and the second alternative generation prompt into the prompt adjustment platform. Through the prompt adjustment platform, adjust the second alternative generation prompt according to the marked second test recommendation information to generate the second generation prompt. The prompt adjustment platform can determine the second test recommendation information that meets the recommendation information requirements and the second test recommendation information that does not meet the recommendation information requirements according to the positive marking and negative marking, and adjust the second alternative generation prompt in the direction of generating the recommendation information that meets the recommendation information requirements, so as to obtain the second generation prompt. The second generation prompt may or may not carry an example of the target recommendation information.

[0094] In one example, the second alternative generation prompt can be used to generate the second test recommendation information, and positive marking and negative marking are performed on the generated second test recommendation information based on the recommendation information requirements. Then, through the prompt adjustment platform, the second alternative generation prompt is adjusted according to the marked second test recommendation information to obtain the second alternative generation prompt after one adjustment. Then, use the second alternative generation prompt after one adjustment to generate the second test recommendation information again, and perform positive marking and negative marking on the second test recommendation information generated again based on the recommendation information requirements. If the quantity ratio of the positively marked second test recommendation information is greater than the ratio threshold, then use the second alternative generation prompt after one adjustment as the second generation prompt. If the quantity ratio of the positively marked second test recommendation information is not greater than the ratio threshold, repeat the above process until the quantity ratio of the positively marked second test recommendation information is greater than the ratio threshold to obtain the second generation prompt.

[0095] In one example, sample recommendation information that meets the requirements of the recommendation information can also be supplemented in the second alternative generation prompt word generated for the first time or the second alternative generation prompt word after a certain adjustment. The second test recommendation information is generated using the second alternative generation prompt word carrying the sample recommendation information, and it is detected whether the proportion of the test recommendation information positively marked in the generated second test recommendation information is greater than the proportion threshold, that is, it is detected whether the second alternative generation prompt word carrying the sample recommendation information can be used as the second generation prompt word. By adding sample recommendation information that meets the requirements of the recommendation information to the second alternative generation prompt word, the effect of quickly adjusting the second alternative generation prompt word is achieved.

[0096] It can be seen that through this embodiment, the second generation prompt word can be accurately generated based on the second alternative generation prompt word and the requirements of the recommendation information by testing and adjusting the second alternative generation prompt word.

[0097] In one embodiment, through the second generative model, content expansion is performed based on the summary and the second dialogue information according to the requirements of the information style, information length, information sentence pattern, and information quantity, and the target recommendation information is generated, including: Through the second generative model, the first core content of the first video data is determined based on the summary, and the second core content of the second video data is determined based on the second dialogue information; Through the second generative model, content expansion is performed based on the first core content and the second core content according to the requirements of the information style, information length, information sentence pattern, and information quantity, and the target recommendation information corresponding to the second video data is generated.

[0098] In this embodiment, after determining the requirements of the information style, information length, information sentence pattern, and information quantity, the first core content of the first video data can be determined through the second generative model based on the summary. The first core content can be the content obtained by summarizing the summary, and is used to further summarize the content of the first video data. The second core content of the second video data can also be determined through the second generative model based on the second dialogue information, and the second core content can be the content summary of the second video data.

[0099] Next, through the second generative model, content expansion is performed based on the first core content and the second core content according to the information style requirements, information length requirements, information sentence pattern requirements, and information quantity requirements, to obtain multiple alternative recommended information. According to the previous introduction, the content of the target recommended information cannot be directly obtained from the second dialogue information and the summary, and needs to be expanded based on the second dialogue information and the summary. Therefore, it is necessary to perform content expansion through the second generative model based on the first core content and the second core content according to the recommended information requirements, and use the expanded recommended information as the alternative recommended information.

[0100] In this embodiment, the name of the short drama or TV drama where the second video data is located, and the introduction of the TV drama or short drama can also be input into the second generative model. Then, the second generative model performs content expansion based on the first core content, the second core content, the drama name, and the drama introduction according to the determined information style requirements, information length requirements, information sentence pattern requirements, and information quantity requirements, to obtain alternative recommended information.

[0101] Finally, based on the alternative recommended information, the target recommended information corresponding to the second video data is generated. The process of generating the target recommended information based on the alternative recommended information can or cannot be generated through the second generative model. In one example, the alternative recommended information containing problem characters can be deleted, and the non-text symbols in the alternative recommended information can be adjusted, and the deleted and adjusted alternative recommended information is used as the target recommended information.

[0102] In this example, for each alternative recommended information, it can be analyzed whether the alternative recommended information contains problem characters, and the problem characters can be preset. If it contains, then this alternative recommended information is deleted. For each alternative recommended information, the non-text symbols in the alternative recommended information can also be adjusted. For example, count each emoji and the number of occurrences of each emoji that appear in all the alternative recommended information, determine the emojis with more occurrences, and when a determined emoji is detected in the alternative recommended information, delete the detected emoji with a certain deletion probability, such as a 50% probability. Finally, the remaining alternative recommended information after deletion and the adjusted alternative recommended information are used as the target recommended information.

[0103] It can be seen that through this embodiment, the first core content of the first video data can be determined based on the summary through the second generative model, the second core content of the second video data can be determined based on the second dialogue information, and content expansion is performed based on the first core content and the second core content according to the information style requirements, information length requirements, information sentence pattern requirements, and information quantity requirements, to generate the target recommended information corresponding to the second video data, thereby achieving the effect of accurately generating the target recommended information.

[0104] In this embodiment, when the second video data is the first episode of a short drama or a TV series, there is no first video data. In this case, the second dialogue information of the second video data can be obtained, and based on the second dialogue information, the target recommendation information corresponding to the second video data is generated through the second generative model. The target recommendation information can be one of the comment information, barrage information, articles or posts recommending the second video data and the video data where it is located.

[0105] Figure 2 The flowchart of the recommendation information generation method provided in another embodiment of the present disclosure is shown as Figure 2 shown. Taking generating comment information for a certain episode of a short drama as the target recommendation information for that episode as an example, in this scenario, an episode of the short drama other than the first five episodes is used as the processing object, and the dialogue information of this single episode of the short drama (i.e., the second dialogue information) is obtained, and in addition, the dialogue information of the first five episodes of this single episode of the short drama (i.e., the first dialogue information) is obtained. Through the first generative model, the preview of this single episode of the short drama is generated based on the dialogue information of the first five episodes. The dialogue information of this single episode of the short drama, the preview of this single episode of the short drama, the name of the short drama, and the pre-obtained short drama introduction are input into the second generative model, and multiple alternative comment information for this single episode of the short drama is generated through the second generative model. And the multiple generated alternative comment information is cleaned to obtain multiple target comment information for this single episode of the short drama, and the target comment information is published to the comment area of this single episode of the short drama.

[0106] In summary, through the above various embodiments, the preview of the second video data and the second dialogue information of the second video data can be automatically generated through the generative model, and the target recommendation information corresponding to the second video data can be automatically generated based on the preview of the second video data and the second dialogue information of the second video data, achieving the effect of automatically generating recommendation information for video data using the generative model and improving the generation efficiency of the recommendation information.

[0107] Figure 3 The structural diagram of the recommendation information generation device provided in an embodiment of the present disclosure is shown as Figure 3 shown. The device includes: An information acquisition unit 31, configured to acquire the first dialogue information of the first video data and the second dialogue information of the second video data; the first video data and the second video data are in the same video data set; in the video data set, the playback order of the first video data is before the playback order of the second video data; A content summary unit 32, configured to generate a content summary of the first video data through the first generative model based on the first dialogue information of the first video data, and use the content summary as the preview of the second video data; An information generation unit 33, configured to generate target recommendation information corresponding to the second video data based on the synopsis of the second video data and the second dialogue information of the second video data through a second generative model.

[0108] Optionally, the information acquisition unit 31 is specifically configured to: perform speech recognition on the first video data to obtain first alternative dialogue information of the first video data, and recognize subtitles and speakers in the video picture of the first video data to obtain second alternative dialogue information of the first video data; use the second alternative dialogue information to adjust the first alternative dialogue information, and use the adjusted first alternative dialogue information as the first dialogue information.

[0109] Optionally, the information acquisition unit 31 is further specifically configured to: determine target dialogue information that is in the second alternative dialogue information and not in the first alternative dialogue information, and supplement the target dialogue information in the first alternative dialogue information; determine contradictory information in the second alternative dialogue information that is contradictory to the first alternative dialogue information, and adjust the first alternative dialogue information based on the contradictory information.

[0110] Optionally, the content summary unit 32 is specifically configured to: obtain a summary generation prompt word corresponding to the content summary; the summary generation prompt word is used to represent the content summary requirements corresponding to the content summary and the content summary example corresponding to the content summary; through the first generative model, summarize the information content of the first dialogue information according to the content summary requirements and the content summary example, to obtain the content summary of the first video data.

[0111] Optionally, the information generation unit 33 is specifically configured to: determine a recommendation scenario for the second video data; the recommendation scenario includes a first scenario or a second scenario; the first scenario is used to comment on the characters and plot in the second video data; the second scenario is used to recommend the second video data to the user; according to the recommendation scenario, generate a first generation prompt word corresponding to the target recommendation information; the first generation prompt word is used to represent the information content requirements corresponding to the target recommendation information; the information content requirements match the recommendation scenario; through the second generative model, expand the content based on the synopsis and the second dialogue information according to the information content requirements, to generate the target recommendation information.

[0112] Optionally, the information generation unit 33 is further specifically configured to: obtain first sample recommendation information matching the recommendation scenario, and generate a first alternative generation prompt word corresponding to the target recommendation information according to the first sample recommendation information; the first alternative generation prompt word is used to prompt the second generative model to generate recommendation information with reference to the first sample recommendation information; based on the information content requirement, test and adjust the first alternative generation prompt word to obtain the first generation prompt word.

[0113] Optionally, the information generation unit 33 is further specifically configured to: obtain first test dialogue information and a first test synopsis, and generate first test recommendation information through the second generative model according to the first alternative generation prompt word, the first test dialogue information, and the first test synopsis; based on the information content requirement, perform positive marking and negative marking on the first test recommendation information to obtain the marked first test recommendation information; adjust the first alternative generation prompt word according to the marked first test recommendation information to generate the first generation prompt word.

[0114] Optionally, the information generation unit 33 is further specifically configured to: extract content from the synopsis according to the information content requirement through the second generative model to obtain a first extraction result, and extract content from the second dialogue information according to the information content requirement to obtain a second extraction result; generate the target recommendation information through the second generative model by performing content expansion according to the first extraction result and the second extraction result.

[0115] Optionally, the information generation unit 33 is specifically configured to: obtain a second generation prompt word corresponding to the target recommendation information; the second generation prompt word is used to represent the recommendation information requirement corresponding to the target recommendation information; the recommendation information requirement includes an information style requirement, an information length requirement, an information sentence pattern requirement, and an information quantity requirement; generate the target recommendation information through the second generative model by performing content expansion based on the synopsis and the second dialogue information according to the information style requirement, the information length requirement, the information sentence pattern requirement, and the information quantity requirement.

[0116] Optionally, the information generation unit 33 is further specifically configured to: obtain second sample recommendation information, and generate a second alternative generation prompt word corresponding to the target recommendation information according to the second sample recommendation information; the second alternative generation prompt word is used to prompt the second generative model to generate recommendation information with reference to the second sample recommendation information; based on the recommendation information requirement, test and adjust the second alternative generation prompt word to obtain the second generation prompt word.

[0117] Optionally, the information generation unit 33 is further specifically configured to: obtain second test dialogue information and a second test synopsis, and through the second generative model, generate second test recommendation information based on the second alternative generation prompt, the second test dialogue information, and the second test synopsis; based on the recommendation information requirement, perform positive marking and negative marking on the second test recommendation information to obtain the marked second test recommendation information; and adjust the second alternative generation prompt according to the marked second test recommendation information to generate the second generation prompt.

[0118] Optionally, the information generation unit 33 is further specifically configured to: through the second generative model, determine the first core content of the first video data based on the synopsis, and determine the second core content of the second video data based on the second dialogue information; through the second generative model, perform content expansion based on the first core content and the second core content according to the information style requirement, the information length requirement, the information sentence pattern requirement, and the information quantity requirement, to generate the target recommendation information corresponding to the second video data.

[0119] In this embodiment, first, obtain the first dialogue information of the first video data and the second dialogue information of the second video data. The first video data and the second video data are in the same video data set, and the playback order of the first video data in the video data set is before that of the second video data. Then, through the first generative model, generate a content summary of the first video data based on the first dialogue information of the first video data, and use the content summary as the synopsis of the second video data. Finally, through the second generative model, generate the target recommendation information corresponding to the second video data based on the synopsis of the second video data and the second dialogue information of the second video data. It can be seen that through this embodiment, the synopsis of the second video data and the second dialogue information of the second video data can be automatically generated by the generative model, and the target recommendation information corresponding to the second video data can be automatically generated based on the synopsis of the second video data and the second dialogue information of the second video data, achieving the effect of automatically generating recommendation information for video data by using the generative model and improving the generation efficiency of the recommendation information.

[0120] The recommendation information generation device in the embodiments of the present disclosure can implement each process of the above-mentioned recommendation information generation method embodiment and achieve the same effects and functions, which will not be repeated here.

[0121] An embodiment of the present disclosure further provides an electronic device. Figure 4 is a schematic structural diagram of the electronic device provided by an embodiment of the present disclosure, as Figure 4As shown, electronic devices can vary significantly due to differences in configuration or performance. They can include one or more processors 401 and a memory 402. One or more applications or data can be stored in the memory 402. Among them, the memory 402 can be short-term storage or persistent storage. The applications stored in the memory 402 can include one or more modules (not shown in the figure), and each module can include a series of computer-executable instructions in the electronic device. Further, the processor 401 can be set to communicate with the memory 402 and execute a series of computer-executable instructions in the memory 402 on the electronic device. The electronic device can also include one or more power supplies 403, one or more wired or wireless network interfaces 404, one or more input or output interfaces 405, one or more keyboards 406, etc.

[0122] In a specific embodiment, the electronic device includes a processor; and a memory configured to store computer-executable instructions that, when executed, cause the processor to implement the following process: Obtain the first dialogue information of the first video data and the second dialogue information of the second video data; the first video data and the second video data are in the same video data set; the playback order of the first video data in the video data set is before the playback order of the second video data; Through a first generative model, based on the first dialogue information of the first video data, generate a content summary of the first video data, and use the content summary as a preview for the second video data; Through a second generative model, based on the preview of the second video data and the second dialogue information of the second video data, generate target recommendation information corresponding to the second video data.

[0123] In this embodiment, first, the first dialogue information of the first video data and the second dialogue information of the second video data are obtained. The first video data and the second video data are in the same video data set, and the playback order of the first video data in the video data set is before that of the second video data. Then, through the first generative model, based on the first dialogue information of the first video data, a content summary of the first video data is generated, and the content summary is used as the preview of the second video data. Finally, through the second generative model, based on the preview of the second video data and the second dialogue information of the second video data, the target recommendation information corresponding to the second video data is generated. It can be seen that through this embodiment, the preview of the second video data and the second dialogue information of the second video data can be automatically generated by the generative model, and the target recommendation information corresponding to the second video data can be automatically generated based on the preview of the second video data and the second dialogue information of the second video data, achieving the effect of automatically generating recommendation information for video data by using the generative model and improving the generation efficiency of the recommendation information.

[0124] The electronic device in the embodiments of the present disclosure can implement each process of the above-mentioned recommendation information generation method embodiment and achieve the same effects and functions, which will not be repeated here.

[0125] Another embodiment of the present disclosure also provides a computer-readable storage medium, which is used to store computer-executable instructions. When the computer-executable instructions are executed by a processor, the following processes are implemented: Obtain the first dialogue information of the first video data and the second dialogue information of the second video data; the first video data and the second video data are in the same video data set; the playback order of the first video data in the video data set is before that of the second video data; Through the first generative model, based on the first dialogue information of the first video data, generate a content summary of the first video data, and use the content summary as the preview of the second video data; Through the second generative model, based on the preview of the second video data and the second dialogue information of the second video data, generate the target recommendation information corresponding to the second video data.

[0126] In this embodiment, first, obtain the first dialogue information of the first video data and the second dialogue information of the second video data. The first video data and the second video data are in the same video data set, and the playback order of the first video data in the video data set is before that of the second video data. Then, through the first generative model, based on the first dialogue information of the first video data, generate the content summary of the first video data, and use the content summary as the preview of the second video data. Finally, through the second generative model, based on the preview of the second video data and the second dialogue information of the second video data, generate the target recommendation information corresponding to the second video data. It can be seen that through this embodiment, the preview of the second video data and the second dialogue information of the second video data can be automatically generated by the generative model, and the target recommendation information corresponding to the second video data can be automatically generated based on the preview of the second video data and the second dialogue information of the second video data, achieving the effect of automatically generating recommendation information for video data by using the generative model and improving the generation efficiency of the recommendation information.

[0127] The computer-readable storage medium in the embodiments of the present disclosure can implement each process of the above-mentioned recommendation information generation method embodiment and achieve the same effects and functions, which will not be repeated here.

[0128] Another embodiment of the present disclosure also provides a computer program product, where the computer program product includes a computer program, and when the computer program is executed by a processor, the following processes are implemented: Obtain the first dialogue information of the first video data and the second dialogue information of the second video data; the first video data and the second video data are in the same video data set; the playback order of the first video data in the video data set is before that of the second video data; Through the first generative model, based on the first dialogue information of the first video data, generate the content summary of the first video data, and use the content summary as the preview of the second video data; Through the second generative model, based on the preview of the second video data and the second dialogue information of the second video data, generate the target recommendation information corresponding to the second video data.

[0129] In this embodiment, first, the first dialogue information of the first video data and the second dialogue information of the second video data are obtained. The first video data and the second video data are in the same video data set, and the playback order of the first video data in the video data set is before that of the second video data. Then, through the first generative model, based on the first dialogue information of the first video data, a content summary of the first video data is generated, and the content summary is used as the preview of the second video data. Finally, through the second generative model, based on the preview of the second video data and the second dialogue information of the second video data, the target recommendation information corresponding to the second video data is generated. It can be seen that through this embodiment, the preview of the second video data and the second dialogue information of the second video data can be automatically generated by the generative model, and the target recommendation information corresponding to the second video data can be automatically generated based on the preview of the second video data and the second dialogue information of the second video data, achieving the effect of automatically generating recommendation information for video data by using the generative model and improving the generation efficiency of the recommendation information.

[0130] The computer program product in the embodiments of the present disclosure can implement each process of the above-mentioned recommendation information generation method embodiment and achieve the same effects and functions, which will not be repeated here.

[0131] In various embodiments of the present disclosure, the computer-readable storage medium includes a read-only memory (ROM for short), a random access memory (RAM for short), a magnetic disk, an optical disc, etc.

[0132] In the 1990s, it was obvious to distinguish whether an improvement to a technology was a hardware improvement (e.g., improvement to the circuit structure of diodes, transistors, switches, etc.) or a software improvement (improvement to the method flow). However, with the development of technology, many improvements to method flows today can be regarded as direct improvements to the hardware circuit structure. Almost all designers obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement to a method flow cannot be implemented with a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user programming the device. Designers can program by themselves to "integrate" a digital system on a piece of PLD, without having to ask a chip manufacturer to design and fabricate a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly implemented using "logic compiler" software, which is similar to the software compiler used in program development and writing. The original code before compilation also has to be written in a specific programming language, which is called a hardware description language (HDL), and there is not only one kind of HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. Currently, the most commonly used are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also be aware that as long as the method flow is slightly logically programmed with the above-mentioned several hardware description languages and programmed into the integrated circuit, it is easy to obtain the hardware circuit that implements the logical method flow.

[0133] The controller can be implemented in any suitable manner. For example, the controller can take the form of, for example, a microprocessor or a processor and a computer-readable medium storing computer-readable program code (such as software or firmware) executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of the controller include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in the form of pure computer-readable program code, it is entirely possible to logically program the method steps to enable the controller to be implemented in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. to achieve the same function. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be regarded as the structures within the hardware component. Or even, the devices for implementing various functions can be regarded as either software modules for implementing the method or structures within the hardware component.

[0134] The systems, devices, modules, or units illustrated in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0135] For the convenience of description, when describing the above devices, they are described separately as various units according to their functions. Of course, when implementing the embodiments of the present disclosure, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0136] Those skilled in the art should understand that one or more embodiments of the present disclosure can be provided as a method, a system, or a computer program product. Therefore, one or more embodiments of the present disclosure can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, one or more embodiments of the present disclosure can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program code.

[0137] This disclosure is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the disclosure. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate a means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0138] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including an instruction means that implements the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0139] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0140] It should also be noted that the term "comprising", "including", or any other variation thereof is intended to cover non-exclusive inclusion, such that a process, method, commodity, or device including a series of elements not only includes those elements but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, commodity, or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, commodity, or device including the said element.

[0141] One or more embodiments of the present disclosure may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. One or more embodiments of the present disclosure may also be practiced in a distributed computing environment where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media including storage devices.

[0142] Each embodiment in the present disclosure is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and reference can be made to the corresponding parts of the method embodiments for the relevant content.

[0143] The above description is only for the embodiments of the present disclosure and is not intended to limit the present disclosure. For those skilled in the art, various changes and modifications can be made to the present disclosure. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included within the scope of the claims of the present disclosure.

Claims

1. A method for generating recommendation information, characterized in that: include: Acquire first conversation information of first video data and second conversation information of second video data; The first video data and the second video data are located in the same video data set; In the video data set, the play order of the first video data is before the play order of the second video data; Generate a content summary of the first video data based on the first conversation information of the first video data by using a first generative model, and use the content summary as a background summary of the second video data; The target recommendation information corresponding to the second video data is generated based on the background summary of the second video data and the second dialogue information of the second video data through a second generative model.

2. The method according to claim 1, characterized in that The step of obtaining the first conversation information of the first video data includes: Performing speech recognition on the first video data to obtain first candidate dialogue information of the first video data, and recognizing subtitles and speakers in a video screen of the first video data to obtain second candidate dialogue information of the first video data; The first candidate dialogue information is adjusted using the second candidate dialogue information, and the adjusted first candidate dialogue information is used as the first dialogue information.

3. The method according to claim 2, characterized in that The adjusting the first candidate dialogue information by using the second candidate dialogue information includes: determining target dialogue information that is in the second candidate dialogue information and is not in the first candidate dialogue information, and supplementing the target dialogue information in the first candidate dialogue information; Conflicting information in the second candidate dialogue information that is inconsistent with the first candidate dialogue information is determined, and the first candidate dialogue information is adjusted based on the conflicting information.

4. The method according to claim 1, characterized in that: The step of generating a content summary of the first video data based on the first dialogue information of the first video data by using the first generative model includes: Obtain summary generation prompt words corresponding to the content summary; the summary generation prompt words are used to indicate content summary requirements corresponding to the content summary and content summary examples corresponding to the content summary; The first generative model is used to summarize the information content of the first dialogue information according to the content summary requirement and the content summary example to obtain the content summary of the first video data.

5. The method according to claim 1, characterized in that The generating, by the second generative model, target recommendation information corresponding to the second video data based on the context summary of the second video data and the second dialogue information of the second video data includes: Determine a recommended scene for the second video data; the recommended scene includes a first scene or a second scene; the first scene is used to comment on the characters and plot in the second video data; the second scene is used to recommend the second video data to the user; According to the recommendation scenario, a first generation prompt word corresponding to the target recommendation information is generated; the first generation prompt word is used to indicate the information content requirement corresponding to the target recommendation information; the information content requirement matches the recommendation scenario; The target recommendation information is generated by performing content expansion based on the previous situation summary and the second dialogue information through the second generative model in accordance with the information content requirement.

6. The method according to claim 5, characterized in that The step of generating a first generation prompt word corresponding to the target recommendation information according to the recommendation scenario includes: Acquire first sample recommendation information matching the recommendation scenario, and generate a first candidate generation prompt word corresponding to the target recommendation information according to the first sample recommendation information; the first candidate generation prompt word is used to prompt the second generative model to generate recommendation information with reference to the first sample recommendation information; Based on the information content requirement, the first candidate generation prompt word is tested and adjusted to obtain the first generation prompt word.

7. The method according to claim 6, characterized in that The step of testing and adjusting the first candidate generated prompt word based on the information content requirement to obtain the first generated prompt word includes: Acquire first test dialogue information and a first test context summary, and generate first test recommendation information according to the first candidate generation prompt word, the first test dialogue information and the first test context summary through the second generative model; Based on the information content requirement, positively mark and negatively mark the first test recommendation information to obtain the marked first test recommendation information; The first candidate generation prompt word is adjusted according to the marked first test recommendation information to generate the first generation prompt word.

8. The method according to claim 5, characterized in that The step of performing content expansion based on the previous situation summary and the second dialogue information according to the information content requirement by the second generative model to generate the target recommendation information includes: By using the second generative model, according to the information content requirement, extracting the content of the previous situation summary to obtain a first extraction result, and, according to the information content requirement, extracting the content of the second dialogue information to obtain a second extraction result; The target recommendation information is generated by performing content expansion according to the first extraction result and the second extraction result through the second generative model.

9. The method according to claim 1, characterized in that: The generating, by the second generative model, target recommendation information corresponding to the second video data based on the context summary of the second video data and the second dialogue information of the second video data includes: Acquire a second generated prompt word corresponding to the target recommended information; the second generated prompt word is used to indicate the recommended information requirement corresponding to the target recommended information; the recommended information requirement includes information style requirement, information length requirement, information sentence requirement and information quantity requirement; Through the second generative model, content expansion is performed based on the previous summary and the second dialogue information in accordance with the information style requirement, the information length requirement, the information sentence requirement and the information quantity requirement to generate the target recommendation information.

10. The method according to claim 9, characterized in that The obtaining of the second generated prompt word corresponding to the target recommendation information includes: Acquire second sample recommendation information, and generate a second candidate generation prompt word corresponding to the target recommendation information according to the second sample recommendation information; the second candidate generation prompt word is used to prompt the second generative model to generate recommendation information with reference to the second sample recommendation information; Based on the recommendation information requirement, the second candidate generation prompt word is tested and adjusted to obtain the second generation prompt word.

11. The method according to claim 10, characterized in that The step of testing and adjusting the second candidate generated prompt word based on the recommendation information requirement to obtain the second generated prompt word includes: Acquire second test dialogue information and a second test context summary, and generate second test recommendation information according to the second candidate generation prompt word, the second test dialogue information and the second test context summary through the second generative model; Based on the recommendation information requirement, positively marking and negatively marking the second test recommendation information to obtain marked second test recommendation information; The second candidate generation prompt word is adjusted according to the marked second test recommendation information to generate the second generation prompt word.

12. The method according to claim 9, characterized in that The step of performing content expansion based on the previous situation summary and the second dialogue information to generate the target recommendation information by the second generative model according to the information style requirement, the information length requirement, the information sentence requirement and the information quantity requirement includes: Determine, by the second generative model, a first core content of the first video data based on the antecedent summary, and determine a second core content of the second video data based on the second dialogue information; Through the second generative model, content expansion is performed based on the first core content and the second core content in accordance with the information style requirements, the information length requirements, the information sentence requirements and the information quantity requirements, to generate the target recommendation information corresponding to the second video data.

13. A device for generating recommendation information, characterized in that: include: An information acquisition unit, used to acquire first conversation information of the first video data and second conversation information of the second video data; The first video data and the second video data are located in the same video data set; in the video data set, the play order of the first video data is located before the play order of the second video data; a content summarizing unit, configured to generate a content summary of the first video data based on the first dialogue information of the first video data by using a first generative model, and use the content summary as a background summary of the second video data; The information generating unit is used to generate target recommendation information corresponding to the second video data based on the background summary of the second video data and the second dialogue information of the second video data through a second generative model.

14. An electronic device, characterized in that: include: processor; as well as, A memory configured to store computer executable instructions, which, when executed, cause the processor to implement the method of any one of claims 1 to 12.

15. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store computer-executable instructions, and the computer-executable instructions implement the method of any one of claims 1 to 12 when executed by a processor.

16. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 12 is implemented.

Images