Language model-based debate information generation method and device, equipment and storage medium

Abstract

The invention discloses a debate information generation method and device based on a language model, equipment and a storage medium, and relates to the technical field of artificial intelligence, the emotional expression ability of debate information is enhanced through a preset language model, so that the output debate information has the emotional enhancement characteristic of each debate stage in expression, and the debate information generation efficiency is improved. The debate information is endowed with more emotions to a certain extent, so that the emotion tension of the debate information in a real debate scene is improved. The method comprises the steps that an operation instruction of generating debate information by a user is received, the operation instruction carries debate information input by the user, and the type of the debate information comprises any one or more of texts, audios and pictures; and based on a preset language model and the dialectical information, generating debate information which corresponds to the dialectical information and has emotional enhancement characteristics of each debate stage.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device, equipment and storage medium for generating debate information based on a language model. Background Art

[0002] In the field of artificial intelligence, language models are based on deep learning and a large amount of training data, and can support a variety of complex language generation and interaction modes. Taking the application of debate scenes as an example, the large language model in the debate scene can help both sides of the debate better understand the topic and find arguments, thereby outputting debate information with higher quality and persuasiveness. However, in real debates, what determines the outcome of the debate is not just the statement of opinions, but often involves emotions, tone and communication skills. Complex emotional changes and excellent communication skills can directly affect the outcome of the debate. The debate information output by the existing large language model is just a statement of opinions, and the expression is usually more mechanical, lacking the emotional tension in real debate scenes, and it is difficult to stimulate emotional resonance or real thinking challenges. Summary of the invention

[0003] In view of this, the present application provides a method, apparatus, device and storage medium for generating debate information based on a language model, the main purpose of which is to solve the problem in the prior art that the debate information output by the large language model is usually a statement of opinion, and the expression is usually more mechanical, which is difficult to arouse emotional resonance or real thinking challenges.

[0004] According to a first aspect of the present application, a method for generating debate information based on a language model is provided, comprising:

[0005] Receive an operation instruction from a user to generate debate information, wherein the operation instruction carries debate topic information input by the user, and the debate topic information type includes any one or more of text, audio, and picture;

[0006] Based on a preset language model and the debate topic information, debate information corresponding to the debate topic information and having characteristics of emotion enhancement at each stage of the debate is generated.

[0007] According to a second aspect of the present application, a debate information generating device based on a language model is provided, comprising:

[0008] A receiving unit, configured to receive an operation instruction from a user to generate debate information, wherein the operation instruction carries debate topic information input by the user, and the debate topic information type includes any one or more of text, audio, and picture;

[0009] A generating unit is used to generate debate information corresponding to the debate topic information and having characteristics of emotion enhancement at each stage of the debate based on a preset language model and the debate topic information.

[0010] According to a third aspect of the present application, a computer device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method described in the first aspect when executing the computer program.

[0011] According to a fourth aspect of the present application, a readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described in the first aspect are implemented.

[0012] By means of the above technical scheme, the present application provides a method and device for generating debate information based on a language model. Compared with the current prior art method of generating debate information based on a large language model, the present application receives an operation instruction from a user to generate debate information, where the operation instruction carries the debate topic information input by the user, and the debate topic information type includes any one or more of text, audio, and picture; based on the preset language model and the debate topic information, the debate information corresponding to the debate topic information is generated with the characteristics of emotional enhancement at each stage of the debate. The whole process enhances the emotional expression ability of the debate information through the preset language model, so that the output debate information has the characteristics of emotional enhancement at each stage of the debate in expression, giving the debate information more emotions to a certain extent, thereby enhancing the emotional tension of the debate information in the real debate scene.

[0013] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0015] Figure 1 It is a flowchart of a method for generating debate information based on a language model provided in an embodiment of the present application;

[0016] Figure 2 It is a flowchart of another method for generating debate information based on a language model provided in an embodiment of the present application;

[0017] Figure 3 It is a flowchart of another method for generating debate information based on a language model provided in an embodiment of the present application;

[0018] Figure 4It is a schematic diagram of a scenario application of a method for generating debate information based on a language model provided in an embodiment of the present application;

[0019] Figure 5 This is a flowchart of another method for generating debate information based on a language model provided in an embodiment of the present application.

[0020] Figure 6 It is a structural schematic diagram of a debate information generating device based on a language model provided in an embodiment of the present application;

[0021] Figure 7 It is a schematic diagram of the device structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0022] Now the content of the present invention will be discussed with reference to several exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and thus implement the content of the present invention, rather than implying any limitation on the scope of the present invention.

[0023] As used herein, the term "including" and variations thereof are to be interpreted as open-ended terms meaning "including but not limited to." The term "based on" is to be interpreted as "based, at least in part, on." The terms "one embodiment" and "an embodiment" are to be interpreted as "at least one embodiment." The term "another embodiment" is to be interpreted as "at least one other embodiment."

[0024] In related technologies, the debate information output by the large language model is merely a statement of opinion, and is usually more mechanical in expression, lacking the emotional tension in a real debate scenario, making it difficult to arouse emotional resonance or real thinking challenges.

[0025] In order to solve this problem, this embodiment provides a method for generating debate information based on a language model, such as Figure 1 As shown, the method comprises the following steps:

[0026] 101. Receive an operation instruction from a user to generate debate information.

[0027] In this embodiment, the debate scene usually provides debate topic information, where the operation instruction carries the debate topic information input by the user, and the debate topic information type includes any one or more of text, audio, and picture. For text-type debate topic information, the debate topic information input by the user can be a piece of text, for audio-type debate topic information, the debate topic information input by the user can be a piece of voice, and for picture-type debate topic information, the debate topic information input by the user can be text debate topic information recognized by AI, or it can be debate topic information that may correspond to the picture understood by AI.

[0028] Correspondingly, the information used for debate output based on the debate topic information is equivalent to the debate information, including the debate information output at each stage of the debate, for example, the debate information can be the views of the pros and cons in the viewpoint output stage, the debate information can be the opening arguments of the pros and cons in the opening argument stage, the debate information can be the content of the pros and cons in the questioning stage, the debate information can be the content of the free debate of the pros and cons in the free debate stage, and the debate information can be the summary of the pros and cons in the summary stage. For example, the debate topic information input by the user is "In the era of big data, our lives have become easier / heavier", then the debate information generated by the debate based on the debate topic information includes the views of the pros and cons, that is, "In the era of big data, our lives have become easier" and "In the era of big data, our lives have become heavier". It can be understood that the debate information can only generate the views of either the pros and cons, or can generate the views of both parties, which is not specifically limited in this embodiment.

[0029] 102. Based on a preset language model and the debate topic information input by the user carried in the operation instruction, generate debate information corresponding to the debate topic information and having characteristics of emotion enhancement at each stage of the debate.

[0030] Among them, the debate information includes any one or more of text, voice, video, picture, and virtual objects with real-time interactive functions. For text-type debate information, the generated debate topic information is a text, and for audio-type debate information, the generated debate information is a voice. The voice can be directly generated by a preset language model, or it can be an audio generated from text to speech (TTS) based on the generated text-type debate information. For the debate information combined with text and voice, the generated debate topic information is a voice clip with text description. Here, the virtual object with real-time interactive function can be a digital human with real-time interactive function. By showing one or more feature combinations of the digital human's lip shape, expression, language, tone and body movements, it can better simulate the complete performance of the debater in the real debate scene, vividly present the debate information, and the presentation effect is more appealing and expressive. The interactive driving data of the digital human can be directly output based on the multimodal large language model, or it can be rendered in real time by the digital human rendering engine according to the digital human rendering driving parameters corresponding to the debate information output by the preset language model, wherein the digital human rendering driving parameters can be generated by the pre-trained AI model based on the debate information, and the rendering result is the interactive display of the digital human audio and video associated with the debate information. The debate information output by the virtual object with real-time interactive function shows emotion enhancement characteristics in each stage of the debate. For example, the debate information output by the virtual object with real-time interactive function expresses the emotion enhancement characteristics of confidence in the argument stage, and the debate information output by the virtual object with real-time interactive function expresses the emotion enhancement characteristics of professionalism in the summary stage.

[0031] In one possible implementation, the debate information includes a virtual object with real-time interactive function. Specifically, the debate information corresponding to the debate topic information is displayed through the virtual object with real-time interactive function. The debate information is generated based on a preset language model and the debate topic information, and has the characteristics of emotion enhancement at each stage of the debate.

[0032] In this embodiment, the preset language model is a language model after the basic language model is adjusted for emotion enhancement. Here, the basic language model can be a general language model, or a language model specially trained for debate scenarios, which is not limited here. The specific emotion enhancement adjustment method is to perform emotion enhancement adjustment through prompt words and / or to perform emotion enhancement adjustment through model fine-tuning. Specifically, the process of emotion enhancement adjustment through prompt words is mainly to set prompt words for each stage of the debate, and to train the basic language model through the prompt words to enhance the emotional expression at each stage of the debate. The process of emotion enhancement adjustment through model fine-tuning is mainly to identify the emotional labels at each stage of the debate, and to fine-tune the basic language model through the emotional labels to enhance the emotional expression at each stage of the debate.

[0033] Taking into account the different types of debate topic information, the debate topic information entered by the user in the web page may be an information type that cannot be recognized by the preset language model. In order to generate more accurate debate topic information, the debate topic information can be pre-processed before it is input into the preset language model, so that the debate topic information is converted into an information type that can be recognized by the preset language model. In an implementable scenario, if the preset language model can recognize text-type debate topic information, the debate topic information is converted into text-type debate topic information and then input into the preset language model. For example, for picture-type debate topic information, the text information in the picture can be obtained through text recognition technology, and the text information in the picture is input into the preset language model as text-type debate topic information. Similarly, for audio-type debate topic information, the text information in the audio can be obtained through speech recognition technology, and the text information in the audio is input into the preset language model as text-type debate topic information.

[0034] Of course, if the preset language model has the function of converting information types, debate topic information of different information types can also be directly input into the preset language model, so that the preset language model can output debate information of corresponding type requirements according to the type requirements of the debate information. At this time, no matter what type of debate topic information is, the type of output debate information can be generated according to user needs. For example, the debate topic information type is a picture type, and the user chooses to generate the debate information type as an audio type. By inputting the debate topic information of the picture type into the preset language model, the audio type debate information can be output accordingly.

[0035] It should be noted that the above information types can be a combination of one or more of text, pictures, and audio, and the debate topic information and / or debate information of the corresponding information types can be transmitted to the client through the following methods, a combination of one or more of page display, voice playback, and virtual object interaction. For text-type debate topic information and / or debate information, it can be transmitted to the client through page display, for picture-type debate topic information and / or debate information, it can be transmitted to the client through page display, and for audio-type debate topic information and / or debate information, it can be transmitted to the client through voice playback and / or virtual object interaction.

[0036] Furthermore, in order to enhance the emotional expression of debate information in all aspects of the debate, it is preferred to transmit the debate information to the client through voice playback and / or virtual object interaction. Since voice and virtual models can more realistically reflect emotional ups and downs and interactions than text, the emotional enhancement characteristics of the debate information can be more easily reflected through voice playback and / or virtual interactive objects, thereby enhancing the emotional expression effect of the debate information at all stages of the debate.

[0037] The method for generating debate information based on a language model provided in the embodiment of the present application is compared with the current prior art method of generating debate information based on a large language model. The present application receives an operation instruction from a user to generate debate information; based on the preset language model and the debate topic information input by the user carried by the operation instruction, generates debate information with emotion enhancement characteristics at each stage of the debate corresponding to the debate topic information, where the debate topic information type includes any one or more of text, audio, and pictures. The whole process enhances the emotional expression ability of the debate information through a preset language model, so that the output debate information has the emotion enhancement characteristics of each stage of the debate in expression, giving the debate information more emotions to a certain extent, thereby enhancing the emotional tension of the debate information in the real debate scene.

[0038] In actual application scenarios, the various stages of a debate include any one or more of the argumentation stage, the questioning stage, the free debate stage, and the summary stage. Each stage of the debate corresponds to different emotion enhancement characteristics. The corresponding relationship between the emotion enhancement characteristics of each stage of the debate includes one or more characteristics of tone, method, expression, quotation, and logical structure. It can be understood that each stage of the debate has a debate order in the debate scenario. Generally speaking, the debate order is: argumentation stage, questioning stage, free debate stage, and summary stage. If the various stages of the debate include the argumentation stage, based on the preset language model and the topic information, the debate information with the emotion enhancement characteristics of the argumentation stage corresponding to the topic information can be generated. If the various stages of the debate include the argumentation stage and the questioning stage, based on the preset language model and the topic information, the debate information with the emotion enhancement characteristics of the argumentation stage and the questioning stage corresponding to the topic information can be generated.

[0039] It should be noted that the debate information in the debate scenario is generated based on the order of argument between the two debaters. That is to say, when the debate information of one party in the current debate stage is known, the debate information of the other party in the current debate stage can be generated based on the debate order. It is impossible to generate debate information across debate stages or multiple times for one party. For example, if the debate information of the affirmative party in the questioning stage is known, the debate information of the negative party in the questioning stage can be generated based on the debate information of the affirmative party in the questioning stage. At this time, the debate information after the questioning stage cannot be generated unless the debate has ended in the argument stage.

[0040] In this embodiment, the corresponding relationship between the emotion enhancement characteristics of each stage of the debate includes any one or more of the following:

[0041] The argumentation stage corresponds to the first emotion enhancement feature. Accordingly, the debate information corresponding to the topic information has one or more of the following features: clear debate viewpoints, clear logical relationships, and confident tone; that is, the argumentation stage requires clear organization, clear viewpoints, and the debate information expresses the confidence of the defense. Accordingly, the debate information output by the preset language model in the argumentation stage is as follows: "Hello, judges and opposing debaters! We firmly believe that artificial intelligence can improve people's employment experience. First of all, with the popularization of artificial intelligence, repetitive and mechanical work will be replaced, and people can be liberated from tedious labor and devote themselves to more creative and valuable work. Secondly, artificial intelligence has brought us new employment opportunities, such as data analysts, AI trainers, etc., and expanded the employment market. In summary, artificial intelligence will not only not reduce employment, but will improve people's employment experience."

[0042] The questioning stage corresponds to the second emotion enhancement feature. Accordingly, the debate information corresponding to the topic information has a strong tone and any one or more of the set emotional expressions, including any one or more of the parallelism, contrast, and rhetorical questions; that is, the questioning stage requires a strong tone, and uses parallelism, contrast, rhetorical questions, etc. to enhance momentum. Accordingly, the debate information output by the preset language model in the questioning stage is as follows: "My opponent, do you think that humans are only suitable for repetitive and mechanical work? Artificial intelligence has replaced boring labor, shouldn't we feel happy? Instead of being bound by work, we should pursue innovation and wisdom! Artificial intelligence has brought new jobs, new skills, and new possibilities. Why should we turn a blind eye? Are we going to refuse to make progress and stagnate?"

[0043] The free debate stage corresponds to the third emotion enhancement feature. Accordingly, the debate information corresponding to the debate topic information has a set verbal expression method, which includes life examples and / or colloquial expressions; that is, the free debate stage requires colloquial expressions and life examples to make people more understanding and empathetic. Accordingly, the example of the debate information output by the preset language model for the free debate stage is as follows: "Think about it, are there any examples around us that make work easier because of the emergence of AI? For example, doctors use AI to assist in diagnosis and find symptoms faster and more accurately; teachers use smart education platforms to make teaching more targeted. Many people have found high-paying jobs by learning AI-related skills. Isn't it because of AI that everyone's employment experience has become better?"

[0044] The summary stage corresponds to the fourth emotion enhancement feature. Accordingly, the debate information corresponding to the topic information has the citation feature of the set field to provide professional debate support. The citation feature here includes text citation and / or key data citation, and the set field is the field associated with the topic information. In other words, the summary stage requires the use of key data and classic citations of famous works to enhance professionalism. Accordingly, the debate information output by the preset language model in the debate stage is as follows: "As shown in the 2020 Future Employment Report, artificial intelligence will create nearly 60 million new jobs in the next five years. At the same time, Microsoft CEO Satya Nadella once said: "Artificial intelligence is not replacing humans, but enhancing human capabilities." Data and authoritative opinions have proved that artificial intelligence is actively changing the job market and improving people's employment experience. We should follow the trend and embrace the future!"

[0045] In actual application scenarios, considering the training process of the preset language model, further, such as Figure 2 As shown, before step 101, the method further includes the following steps:

[0046] 201. Obtain sample information of debate scenarios.

[0047] 202. Constructing emotional instructions for enhancing the model's ability to express emotions during the debate based on the emotional labels of each stage of the debate.

[0048] 203. Fine-tune the basic language model using the emotion instruction and the sample information to generate a preset language model.

[0049] The sample information includes debate topic sample information and debate sample information corresponding to each stage of the debate. The debate sample information may be debate content provided by a debate competition in a debate scenario, or may be debate content generated based on a model.

[0050] Specifically, the sample information of the debate scene is the debate information generated by the debate at each stage of the debate based on the provision of the debate topic information. Since the debaters are interactive, the debate information is also recorded in an interactive form. That is to say, in each stage of the debate, after one party provides the debate information, the other party provides the debate information until the current debate stage ends and the next debate stage begins, and the debaters continue to execute the interactive process of debate information.

[0051] Considering the emotional ups and downs of debate information in real debate scenarios, the debate sample information at different stages of the debate usually has different emotional expressions. Figure 3 As shown, after step 201, the method further includes the following steps:

[0052] 301. Perform emotion recognition on the debate sample information at each stage of the debate based on a preset emotion recognition model to obtain emotion features contained in the debate sample information.

[0053] 302. Correlate the debate sample information at each stage of the debate with the emotional features contained in the corresponding debate sample information to obtain the emotional features corresponding to each stage of the debate.

[0054] In this embodiment, the emotional features contained in each stage of the debate can be reflected by the text structure and tone with emotional expression. For the same emotional features, different text structures will change the emotional expression of the debate information. For example, the parallelism method can emphasize the momentum expression of the debate information, and the interrogative sentence can emphasize the uncertainty expression of the debate topic information. For the same emotional features, different voices / intonations can change the emotional expression of the debate information. For example, the rhetorical tone can emphasize the momentum expression of the debate information, and the exclamatory tone can emphasize the emotional expression of the debate information. Specifically, based on the text structure and tone contained in the debate sample information, the debate sample information at each stage of the debate can be emotionally recognized by a preset emotion recognition model to obtain the emotional features contained in the debate sample information. For the target debate sample information containing the text structure and tone with emotional expression, the emotional features contained in the debate sample information can be accurately identified.

[0055] It can be understood that in order to enable the preset language model to have the ability to express emotions in a real debate scenario, the debate sample information is labeled according to the emotional features corresponding to each stage of the debate, and the debate sample information at each stage of the debate can be associated with the emotional features contained in the corresponding debate sample information.

[0056] Specifically, after associating the debate sample information at each stage of the debate with the emotional features contained in the corresponding debate sample information, the emotional features corresponding to each stage of the debate can be described by an emotional mapping table, and the emotional mapping table can be in the following form:

[0057] Emotional characteristics corresponding to the argumentation stage: Confidence (clear logic and clear views)

[0058] Emotional characteristics corresponding to the debate stage - passionate (strong tone, using parallelism, contrast, rhetorical questions, etc. to enhance momentum)

[0059] Emotional characteristics of free debate - empathy (colloquial expression, daily examples, so that people can understand and empathize better)

[0060] Emotional characteristics corresponding to the summary stage - professionalism (key data and quotations from famous classics enhance professionalism and make people more convincing)

[0061] In actual application scenarios, the debate information generation method based on language models can be applied to debate platforms. Based on the debate topic information input by the user, the debate platform can generate a debate generation task according to the debate topic information. Through the preset language model, the debate generation task can be disassembled and iteratively analyzed in the order of the debate time to generate multiple rounds of debate information. Each round of debate information is interactive and can be refuted based on the debate information of the previous round.

[0062] In actual application scenarios, the user's operation instruction to generate debate information can be triggered by web page behavior. For example, after the user uploads the debate topic information to the web page, the user's operation instruction to generate debate information is triggered by clicking. It can also be automatically triggered by the background server. For example, after the user uploads the debate topic information to the web page, the background server automatically triggers the user's operation instruction to generate debate information. The scenario application of the specific debate information generation method is shown in 4. Figure 4 The debate platform can receive the debate topic information input by the user, and based on the debate topic information, generate debate information with emotional enhancement characteristics at each stage of the debate according to the debate time sequence based on the preset language model. The specific process of generating debate information based on the preset language model according to the debate time sequence is as follows: Figure 5 As shown, Figure 5 On the basis of the debate topic information, the views of both sides of the debate, the opening arguments of both sides of the debate, the content of the debate, the content of the free debate of both sides, and the summary of both sides of the debate are generated in the order of the debate time based on the preset language model. Furthermore, in order to make the preset language model generate debate information with emotion enhancement features, the prompt information of each stage of the debate can be extracted from the debate requirements corresponding to each stage of the debate, and the prompt information of each stage of the debate is used as the control instruction of emotion expression. The prompt information and the debate topic information are input into the preset language model together, and the debate information with emotion enhancement features of each stage of the debate corresponding to the debate topic information is generated through the preset language model.

[0063] Specifically, based on the input of the debate topic information, the views of both sides of the debate are generated first according to the time sequence of the debate based on the preset language model. It should be noted that the views of both sides of the debate are generated synchronously after the debate topic information is input into the preset language model, and there is no order of precedence. Accordingly, the following prompt information can be set for the views of both sides of the debate:

[0064] Please divide this topic information into two topics: affirmative and negative. Topic: "{topic}"

[0065] Height Line:

[0066] 1. Output only the topic, without any details or explanation

[0067] 2. Output format:

[0068] Affirmation: xxx (limited to 20 characters)

[0069] Negative number: xxx (limited to 20 characters)

[0070] Specifically, based on the topic information and the views of both parties of the debate, the opening arguments of both parties of the debate are generated in the order of the debate time based on the preset language model. It should be noted that the opening arguments of both parties of the debate are generated separately after the topic information and the views of one party of the debate are input into the preset language model, and there is no order of precedence, that is, the opening argument of one party of the debate does not affect the opening argument of the other party of the debate. For the opening argument of the affirmative side, the topic information and the views of the affirmative side of the debate can be input into the preset language model to generate the opening argument of the affirmative side. Accordingly, the following prompt information can be set for the opening argument of the affirmative side:

[0071] Please be the affirmative party in the debate. Topic: "{topic}"

[0072] Your position is affirmative.

[0073] Now, please give your opening argument as the affirmative side.

[0074] Require:

[0075] 1. Your opening statement should state your core point.

[0076] 2. The word count should not exceed 100 words.

[0077] Similarly, for the opening argument of the opposing party, the topic information and the opposing party's viewpoints can be input into the preset language model to generate the opening argument of the opposing party. Accordingly, the following prompt information can be set for the opening argument of the opposing party:

[0078] Please be the opponent of the debate. Topic: "{topic}"

[0079] Your position is negative.

[0080] Now, please give your opening argument as the opposing party.

[0081] Require:

[0082] 1. Your opening statement should state your core point.

[0083] 2. The word count should not exceed 100 words.

[0084] Specifically, based on the topic information, the viewpoints of both parties and the opening arguments of both parties, the content of the debate between the two parties is generated according to the time sequence of the debate based on the preset language model. It should be noted that the content of the debate between the two parties is generated separately after the topic information, the viewpoints of one party and the opening arguments of both parties are input into the preset language model, and there is no order of precedence, that is, the content of the debate by one party does not affect the content of the debate by the other party. For the content of the affirmative side's questioning, the topic information, the viewpoints of the affirmative side and the opening arguments of both parties can be input into the preset language model to generate the content of the affirmative side's questioning. Accordingly, the following form of prompt information can be set for the content of the affirmative side's questioning:

[0085] Please be the affirmative party in the debate. Topic: "{topic}"

[0086] Your position is affirmative.

[0087] Your opening statement is: {aff_opening}

[0088] Now, please give the following rebuttal to the opening argument of the opposing side of the debate: {neg_opening}

[0089] Require:

[0090] 1. Specific rebuttal to the debate opponent's opening argument. Be direct and don't joke.

[0091] 2. The word count should not exceed 60 words.

[0092] Similarly, for the content of the opposing side's questioning, the topic information, the opposing side's viewpoints, and the opening arguments of both sides of the debate can be input into the preset language model to generate the content of the opposing side's questioning. Accordingly, the following prompt information can be set for the content of the opposing side's questioning:

[0093] Please be the opponent of the debate. Topic: "{topic}"

[0094] Your position is negative.

[0095] Your opening statement is: {neg_opening}

[0096] Now, please make the following rebuttal to the affirmative's opening statement: {aff_opening}

[0097] Require:

[0098] 1. Specific rebuttal to the debate's opening argument. Be direct and don't joke.

[0099] 2. The word count should not exceed 60 words.

[0100] Specifically, based on the topic information, the viewpoints of both parties, the opening arguments of both parties, and the content of the debate, the free debate content of both parties is generated according to the time sequence of the debate based on the preset language model. It should be noted that the free debate content of both parties is generated after the topic information, the viewpoints of one party, the opening theories of both parties, and the content of the debate are input into the preset language model. Since the free debate of both parties needs to refute the statement of the other party based on the statement of one party, it means that the free debate of both parties has a sequence, that is, the free debate content of one party affects the free debate content of the other party. For the free debate content of the affirmative side, the topic information, the viewpoints of the affirmative side, the opening theories of both parties, and the content of the debate are input into the preset language model to generate the free debate content of the affirmative side. Accordingly, the following prompt information can be set for the free debate content of the affirmative side:

[0101] Please be the affirmative party in the debate. Topic: "{topic}"

[0102] Your position is affirmative.

[0103] In the previous stage, both sides made the following opening arguments and rebuttals:

[0104] Opening statement of the affirmative side: {aff_Opening}

[0105] Opening statement of the opposing side: {neg_Opening}

[0106] Affirmative rebuttal: {aff_rebutal}

[0107] Negative rebuttal: {neg_rebutal}

[0108] Now, as the affirmative debater, give your first free speech. You can respond to the other side's arguments or bring up new points to strengthen your position.

[0109] Require:

[0110] 1. Keep your speech brief and to the point.

[0111] 2. The word count should not exceed 40 words.

[0112] Similarly, for the content of the free debate of the opposing party, the topic information, the viewpoints of the opposing party, the opening theories of both sides of the debate, the content of the debate of both sides, and the content of the free debate of the affirmative party can be input into the preset language model to generate the content of the free debate of the opposing party. Accordingly, the following prompt information can be set for the content of the free debate of the opposing party:

[0113] Please be the opponent of the debate. Topic: "{topic}"

[0114] Your position is negative.

[0115] In the previous stage, both sides made the following opening arguments and rebuttals:

[0116] Opening statement of the affirmative side: {aff_Opening}

[0117] Opening statement of the opposing side: {neg_Opening}

[0118] Affirmative rebuttal: {aff_rebutal}

[0119] Negative rebuttal: {neg_rebutal}

[0120] Content of the affirmative side free debate: {aff_free1}

[0121] Now, as the opponent, give your first free speech. You can respond to the opponent's arguments or bring up new points to strengthen your position.

[0122] Require:

[0123] 1. Keep your speech brief and to the point.

[0124] 2. The word count should not exceed 40 words.

[0125] Specifically, based on the topic information, the viewpoints of both parties, the opening arguments of both parties, the content of the debate and the free debate of both parties, the summaries of both parties are generated in the order of the debate time based on the preset language model. It should be noted that the summaries of both parties are generated by inputting the topic information, the viewpoints of one party, the opening theories of both parties, the content of the debate and the free debate of both parties into the preset language model. Since the summaries of both parties need to refute the statements of the other party based on the statements of one party, it means that the summaries of both parties have a sequence, that is, the summary of one party affects the summary of the other party. For the affirmative summary, the topic information, the viewpoints of the affirmative party, the opening theories of both parties, the content of the debate and the free debate of both parties can be input into the preset language model to generate the affirmative summary. Accordingly, the following prompt information can be set for the affirmative summary:

[0126] Please be the affirmative party in the debate. Topic: "{topic}"

[0127] Your position is affirmative.

[0128] In the previous stage, both sides made the following opening arguments and rebuttals:

[0129] Opening statement of the affirmative side: {aff_Opening}

[0130] Opening statement of the opposing side: {neg_Opening}

[0131] Affirmative rebuttal: {aff_rebutal}

[0132] Negative rebuttal: {neg_rebutal}

[0133] Content of the affirmative side free debate: {aff_free1}

[0134] Content of the negative free debate: {neg_free1}

[0135] Now, as the affirmative side of the debate, please express your affirmativeness in your summary. You need to first refute the negative side of the core argument, then emphasize the core argument and prove your own point.

[0136] Require:

[0137] 1. Keep your speech brief and to the point.

[0138] 2. The word count should not exceed 50 words.

[0139] Similarly, for the summary of the opposing party, the topic information, the opposing party's viewpoints, the opening theories of both sides, the content of the debate, the free debate content of both sides, and the summary of the affirmative party can be input into the preset language model to generate the summary of the opposing party. Accordingly, the following prompt information can be set for the summary of the opposing party:

[0140] Please be the opponent of the debate. Topic: "{topic}"

[0141] Your position is negative.

[0142] In the previous stage, both sides made the following opening arguments and rebuttals:

[0143] Opening statement of the affirmative side: {aff_Opening}

[0144] Opening statement of the opposing side: {neg_Opening}

[0145] Affirmative rebuttal: {aff_rebutal}

[0146] Negative rebuttal: {neg_rebutal}

[0147] Content of the affirmative side free debate: {aff_free1}

[0148] Content of the negative free debate: {neg_free1}

[0149] Affirmative Conclusion: {aff_Closing}

[0150] Now, as the affirmative side of the debate, please state the negative in your summary. You need to first refute the positive side of the core argument, then emphasize the core argument and prove your own point.

[0151] 1. Keep your speech brief and to the point.

[0152] 2. The word count should not exceed 50 words.

[0153] In actual application scenarios, the basic language model relies on statistical generation of sentences when automatically generating debate opinions based on debate topic information, lacking a deep understanding of debate logic, philosophy or other disciplines. Although a large amount of content can be quickly generated, it still has defects in in-depth logical reasoning and complex argumentation, especially when it needs to cross knowledge fields or rigorous argumentation. This makes the debate opinions output by the user through the basic language model not coherent or logical, resulting in an experience that is not real or deep enough. As an improvement method of a preset language model, a special logical reasoning algorithm, such as symbolic reasoning or a rule-based system, can be integrated into the preset language model of this embodiment to assist the preset language model in better understanding and constructing complex debate structures through a logical reasoning algorithm. In this way, by adding a reasoning module to the basic language model, the reasoning module can better capture the causal relationship and logical connection in the debate information. As another improvement method of the preset language model, more high-quality debate data can also be introduced into the preset language model of this embodiment, especially debate data with rigorous argumentation fields, such as philosophy, law, etc., to enhance the ability of the preset language model to generate deep arguments through high-quality debate data.

[0154] In actual application scenarios, although the basic language model can generate appropriate debate information according to the currently input debate topic information, in the long debate process, the basic language model has certain memory limitations when processing context, and the corresponding context window length is limited to a specific number of words. If it exceeds the limit, the previous debate information will be lost, which makes it difficult for the basic language model to remember previous arguments and discussion content, and it is easy to generate repeated or contradictory debate information, so that the basic language model cannot effectively track the context in a long debate scenario, resulting in a decrease in the continuity and consistency of the debate, affecting the overall interactive experience. As an improved method of a preset language model, a long-short memory module and a class persistent module can be introduced into the preset language model of this embodiment to ensure that the preset language model can retain and recall the previous content at each stage of the debate, thereby improving the coherence of the debate information. As another improved method of the preset language model, the core arguments in the debate can also be summarized in real time, and the core arguments are regularly generated by the core arguments. The core argument summary allows the preset language model of this embodiment to review the previous discussion at each stage of the debate according to the core argument summary to ensure the consistency of the debate information context. As another way to improve the preset language model, a special multi-round conversation model is built for long debate scenarios. By combining the multi-round conversation model with the preset language model, long sequences can be effectively processed, and a memory library can be established in different rounds to improve the fluency of users in long debates.

[0155] In practical application scenarios, due to the difficulty of accurately simulating complex emotional changes and subtle interpersonal communication skills in basic language models, lacking emotional recognition and appropriate feedback mechanisms in expression, the debate information automatically generated based on the debate topic information usually lacks emotional tension, making the interactive process of the debate scene more mechanized. As an improved method of a preset language model, the tone and response mode of the debate information can be adjusted using emotion generation technology on the basis of the output of the debate information by the preset language model, so that the debate interactive process is more real and vivid. As an improved method of another preset language model, multiple debate style modules can be designed in the preset language model of the present embodiment, and different debate roles or tone styles (such as strong debaters, gentle debaters) are simulated by different debate style modules, so that debaters can choose a favorite style to enhance the interactive experience. As an improved method of another preset language model, the preset language model of the present embodiment can be combined with scenario modeling, and the social cues and non-explicit information in the dialogue can be understood by the preset language model after scenario modeling, so as to enhance the interactive skills of the preset language model in the debate process.

[0156] In actual application scenarios, personalized restrictions need to be based on the accumulation of user portraits and long-term interaction data, and the basic language model cannot effectively customize the debate styles and needs of different users. Considering that different users have different debate styles and points of interest, the basic language model cannot be dynamically adjusted, resulting in a poor debate interaction experience. As an improved method of a preset language model, a personalized user portrait can be generated for each user by analyzing the user's historical debate data, and the preset language model in this embodiment is trained based on the personalized user portrait, so that the trained preset language model can generate customized debate information based on the debate style, points of interest and knowledge level of different users. As another improved method of the preset language model, it is allowed to adjust the debate parameters of the preset language model in this embodiment, and the debate parameters include one or more combinations of debate depth, debate tone, debate speed, and debate emotional expression. The preset language model after parameter adjustment can better adapt to the user's personal needs. As another improved method of the preset language model, by collecting user feedback evaluations, the performance of the preset language model in this embodiment is continuously adjusted and optimized according to the user's feedback evaluation, and a user feedback closed loop is established to help the system better understand user needs and automatically perform personalized optimization through the user feedback closed loop.

[0157] In actual application scenarios, the basic language model is usually trained based on historical debate data, and it is difficult to obtain the latest debate information in real time. In particular, in debate scenarios with high timeliness requirements, the basic language model usually has a lag and cannot quickly update the corresponding debate information, especially in debates related to current affairs or the latest research. When the debate scenario involves current affairs or emerging topics, it is difficult for the basic language model to give accurate debate information, which reduces the reference value of the debate information. As an improved method of a preset language model, the latest debate information can be obtained by accessing a real-time knowledge base or an Internet search function, so that the preset language model in this embodiment can be adjusted by the latest debate information, thereby ensuring that the debate information generated by the preset language model can be applied to the latest current affairs data. As another improved method of the preset language model, the ability of continuous learning can be added to the preset language model in this embodiment, and the knowledge base can be quickly updated when the latest debate information appears through an incremental learning mechanism, thereby continuously improving the knowledge base docked by the preset language model. As another improved method of the preset language model, when the debate information needs to be updated, the preset language model in this embodiment is connected to an external database through an interface, such as a news database, a latest academic paper library, etc., to supplement the debate information update content through external data.

[0158] In actual application scenarios, the basic language model has limited understanding and reasoning capabilities for complex concepts, and is prone to lack of in-depth perspectives. For sensitive topics such as philosophy and theory, the debate information output by the basic language model is usually superficial and cannot deeply explore the complexity and multi-angle nature of the debate. This makes users feel that the debate information is just repeating existing common arguments when debating complex topics, and cannot provide new or profound insights. As an improvement method for the preset language model, the preset language model of this embodiment can be fine-tuned using specific domain data for different complex topics, and multiple domain expert sub-models can be constructed. The associated domain expert sub-models are dynamically called to provide in-depth arguments during the debate. As another improvement method for the preset language model, a multi-model collaborative structure can be designed, and different domain models can work together during the debate process to ensure that under complex topics, the preset language model in this embodiment can integrate knowledge from multiple fields to provide more comprehensive debate information. As another way to improve the preset language model, it can be combined with logic programming or a special reasoning engine, driven by logic programming and a special reasoning logic engine, so that the preset language model in this embodiment can build a more rigorous reasoning chain during the debate on complex topics, thereby effectively generating rebuttal debate information.

[0159] In actual application scenarios, since the training data may contain bias, the debate information generated by the basic language model may contain implicit bias or unethical rounds. However, although the filtering mechanism has been used to reduce the possibility of biased debate information output by the basic language model, the basic language model still cannot completely avoid learning invisible bias from the training data during the training process, which makes the debate information generated by the basic language model have biased content in terms of gender discrimination and race, which not only destroys the user experience, but also causes discomfort and controversy in the debate information. As an improvement method of the preset language model, a bias detection module can be introduced in the process of the preset language model generating debate information in this embodiment. The bias detection module uses a special algorithm to review the generated debate information to filter out potential biased content. As another improvement method of the preset language model, a variety of training data can be constructed, which covers different cultures, genders, races and backgrounds, etc. The preset language model in this embodiment is trained through a variety of training data, thereby reducing the possibility of absorbing bias during the training process. As another improvement method of the preset language model, before the preset language model generates debate information in this embodiment, an ethical review mechanism can be added to detect whether there are ethical issues in the debate information, and automatically adjust or correct it.

[0160] Further, as Figure 1-Figure 5 The specific implementation of the method, the embodiment of the present application provides a debate information generation device based on a language model, such as Figure 6As shown, the device includes: a receiving unit 61 and a generating unit 62.

[0161] A receiving unit 61 is used to receive an operation instruction of a user to generate debate information, wherein the operation instruction carries debate topic information input by the user, and the debate topic information type includes any one or more of text, audio, and picture;

[0162] The generating unit 62 is used to generate debate information corresponding to the debate topic information and having the characteristics of emotion enhancement at each stage of the debate based on a preset language model and the debate topic information.

[0163] The apparatus for generating debate information based on a language model provided by an embodiment of the present invention is different from the current prior art method of generating debate information based on a large language model. The present application receives an operation instruction from a user to generate debate information; based on a preset language model and the debate topic information input by the user carried by the operation instruction, generates debate information with emotion enhancement characteristics at each stage of the debate corresponding to the debate topic information, where the debate topic information type includes any one or more of text, audio, and picture. The entire process enhances the emotional expression ability of the debate information through a preset language model, so that the output debate information has the emotion enhancement characteristics of each stage of the debate in expression, giving the debate information more emotions to a certain extent, thereby enhancing the emotional tension of the debate information in a real debate scene.

[0164] In actual application scenarios, the preset language model is a language model that has been adjusted for emotion enhancement.

[0165] In actual application scenarios, the emotion enhancement adjustment method is to perform emotion enhancement adjustment through prompt words and / or to perform emotion enhancement adjustment through model fine-tuning.

[0166] In actual application scenarios, the debate stages include any one or more of the argumentation stage, the questioning stage, the free debate stage, and the summary stage.

[0167] In actual application scenarios, each stage of the debate corresponds to different emotion enhancement characteristics, and the corresponding relationship between the emotion enhancement characteristics of each stage of the debate includes one or more characteristics of tone, method, expression, quotation, and logical structure.

[0168] In actual application scenarios, the corresponding relationship between the emotion enhancement characteristics of each stage of the debate includes any one or more of the following:

[0169] The argumentation stage corresponds to the first emotion enhancement feature, and accordingly, the debate information corresponding to the debate topic information has one or more of the following features: clear debate viewpoints, clear logical relationships, and a confident tone;

[0170] The questioning and debate stage corresponds to the second emotion enhancement feature. Accordingly, the debate information corresponding to the debate topic information has any one or more of the following features: a strong tone and a set emotional expression method. The emotional expression method includes any one or more of a parallelism method, a contrast method, and a rhetorical question method.

[0171] The free debate stage corresponds to the third emotion enhancement feature. Accordingly, the debate information corresponding to the debate topic information has a set verbal expression method, and the verbal expression method includes a life-like example method and / or a colloquial expression method;

[0172] The summary stage corresponds to the fourth emotion enhancement feature. Accordingly, the debate information corresponding to the topic information has a citation feature of a set field, and the citation feature includes text citation and / or key data citation. The set field is a field associated with the topic information.

[0173] In actual application scenarios, the debate information includes any one or more of text, voice, video, picture, and virtual objects with real-time interactive functions.

[0174] In actual application scenarios, the virtual object with real-time interactive function is generated based on a language model with multimodal characteristics, and the debate information output by the virtual object with real-time interactive function shows emotion enhancement characteristics at each stage of the debate.

[0175] In actual application scenarios, the debate information includes a virtual object with real-time interactive function. Specifically, the debate information corresponding to the debate topic information is displayed through the virtual object with real-time interactive function. The debate information is generated based on a preset language model and the debate topic information, and has the characteristics of emotional enhancement at each stage of the debate.

[0176] In an actual application scenario, the device further includes:

[0177] An acquisition unit, configured to acquire sample information of a debate scene before receiving an operation instruction from a user to generate debate information, wherein the sample information includes sample information of a debate topic and sample information of debates at various stages of the debate corresponding to the sample information of the debate topic;

[0178] A construction unit, used for constructing emotional instructions for enhancing the emotional expression ability of the model during the debate according to the emotional labels of each stage of the debate;

[0179] A fine-tuning unit is used to use the emotion instruction and the sample information to perform model fine-tuning on the basic language model to generate a preset language model.

[0180] In an actual application scenario, the device further includes:

[0181] an identification unit, configured to, after acquiring the sample information of the debate scene, perform emotion recognition on the debate sample information of each stage of the debate based on a preset emotion recognition model, so as to obtain the emotion features contained in the debate sample information;

[0182] The association unit is used to associate the debate sample information of each stage of the debate with the emotional features contained in the corresponding debate sample information to obtain the emotional features corresponding to each stage of the debate.

[0183] It should be noted that for other corresponding descriptions of the functional units involved in the debate information generation device based on a language model provided in this embodiment, reference can be made to Figure 1-Figure 5 The corresponding description in will not be repeated here.

[0184] Based on the above Figure 1-Figure 5 The method shown in the embodiment of the present application accordingly provides a storage medium on which a computer program is stored, and when the program is executed by a processor, the above-mentioned Figure 1-Figure 6 The debate information generation method based on language model is shown.

[0185] Based on this understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present application.

[0186] Based on the above Figure 1-Figure 5 The method shown, and Figure 6 In order to achieve the above-mentioned purpose, the embodiment of the present application also provides a physical device for generating debate information based on a language model, which can be a computer, a smart phone, a tablet computer, a smart watch, a server, or a network device, etc. The physical device includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to achieve the above-mentioned Figure 1-Figure 5 The debate information generation method based on language model is shown.

[0187] Optionally, the physical device may also include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a WI-FI module, etc. The user interface may include a display, an input unit such as a keyboard, etc., and the optional user interface may also include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (such as a WI-FI interface), etc.

[0188] In an exemplary embodiment, see Figure 7 The physical device includes a communication bus, a processor, a memory and a communication interface, and may also include an input / output interface and a display device, wherein each functional unit can communicate with each other through the bus. The memory stores a computer program, and the processor is used to execute the program stored in the memory and execute the method for generating debate information based on the language model in the above embodiment.

[0189] Those skilled in the art will appreciate that the physical device structure for generating debate information based on a language model provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or a combination of certain components, or different arrangements of components.

[0190] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the entity device for the above-mentioned language model-based debate information generation and processing, and supports the operation of the information processing program and other software and / or programs. The network communication module is used to realize the communication between the components inside the storage medium, and the communication with other hardware and software in the information processing entity device.

[0191] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present application can be implemented by means of software plus the necessary general hardware platform, or by hardware. By applying the technical solution of the present application, compared with the current existing methods, the present application enhances the emotional expression ability of the debate information by presetting the language model, so that the output debate information has the emotional enhancement characteristics of each stage of the debate in expression, and to a certain extent gives the debate information more emotions, thereby enhancing the emotional tension of the debate information in the real debate scene.

[0192] Those skilled in the art will appreciate that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily necessary for implementing the present application. Those skilled in the art will appreciate that the modules in the devices in the implementation scenario can be distributed in the devices of the implementation scenario according to the description of the implementation scenario, or can be changed accordingly and located in one or more devices different from the present implementation scenario. The modules of the above-mentioned implementation scenario can be combined into one module, or can be further split into multiple submodules.

[0193] The above serial numbers of this application are only for description and do not represent the advantages and disadvantages of the implementation scenarios. The above disclosure is only a few specific implementation scenarios of this application, but this application is not limited to them, and any changes that can be thought of by technicians in this field should fall within the scope of protection of this application.

Claims

1. A method for generating debate information based on a language model, characterized in that: include: Receive an operation instruction from a user to generate debate information, wherein the operation instruction carries debate topic information input by the user, and the debate topic information type includes any one or more of text, audio, and picture; Based on a preset language model and the debate topic information, debate information corresponding to the debate topic information and having characteristics of emotion enhancement at each stage of the debate is generated.

2. The method according to claim 1, characterized in that The preset language model is a language model that has been adjusted through emotion enhancement.

3. The method according to claim 2, characterized in that The emotion enhancement adjustment method is to perform emotion enhancement adjustment through prompt words and / or perform emotion enhancement adjustment through model fine-tuning.

4. The method according to claim 1, characterized in that: The debate stages include any one or more of the argumentation stage, the questioning stage, the free debate stage and the summary stage.

5. The method according to claim 4, characterized in that Each stage of the debate corresponds to different emotion enhancement characteristics, and the corresponding relationship between the emotion enhancement characteristics of each stage of the debate includes one or more characteristics of tone, method, expression, quotation, and logical structure.

6. The method according to claim 5, characterized in that The corresponding relationship between the emotion enhancement characteristics of each stage of the debate includes any one or more of the following: The argumentation stage corresponds to the first emotion enhancement feature, and accordingly, the debate information corresponding to the debate topic information has one or more of the following features: clear debate viewpoints, clear logical relationships, and a confident tone; The questioning and debate stage corresponds to the second emotion enhancement feature. Accordingly, the debate information corresponding to the debate topic information has any one or more of the following features: a strong tone and a set emotional expression method. The emotional expression method includes any one or more of a parallelism method, a contrast method, and a rhetorical question method. The free debate stage corresponds to the third emotion enhancement feature. Accordingly, the debate information corresponding to the debate topic information has a set verbal expression method, and the verbal expression method includes a life-like example method and / or a colloquial expression method; The summary stage corresponds to the fourth emotion enhancement feature. Accordingly, the debate information corresponding to the topic information has a citation feature of a set field, and the citation feature includes text citation and / or key data citation. The set field is a field associated with the topic information.

7. The method according to claim 1, characterized in that The debate information includes any one or more of text, voice, video, picture, and virtual objects with real-time interactive functions.

8. The method according to claim 1 or 7, characterized in that: The debate information includes a virtual object with real-time interactive function. Specifically, the debate information corresponding to the debate topic information is displayed through the virtual object with real-time interactive function. The debate information is generated based on a preset language model and the debate topic information, and has the characteristics of emotion enhancement at each stage of the debate.

9. The method according to any one of claims 1 to 8, characterized in that Before receiving the operation instruction of the user to generate debate information, the method further includes: Acquire sample information of a debate scene, the sample information including sample information of a debate topic and sample information of debates at various stages of the debate corresponding to the sample information of the debate topic; According to the emotion labels of each stage of the debate, constructing emotion instructions for enhancing the model's ability to express emotions during the debate; The basic language model is fine-tuned using the emotion instruction and the sample information to generate a preset language model.

10. The method according to claim 9, characterized in that After acquiring sample information of the debate scene, the method further includes: Performing emotion recognition on the debate sample information at each stage of the debate based on a preset emotion recognition model to obtain the emotion features contained in the debate sample information; The debate sample information of each stage of the debate is associated with the emotional features contained in the corresponding debate sample information to obtain the emotional features corresponding to each stage of the debate.

11. A debate information generating device based on a language model, characterized in that: include: A receiving unit, configured to receive an operation instruction from a user to generate debate information, wherein the operation instruction carries debate topic information input by the user, and the debate topic information type includes any one or more of text, audio, and picture; A generating unit is used to generate debate information corresponding to the debate topic information and having characteristics of emotion enhancement at each stage of the debate based on a preset language model and the debate topic information.

12. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method for generating debate information based on a language model according to any one of claims 1 to 10 are implemented.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for generating debate information based on a language model according to any one of claims 1 to 10 are implemented.

Images