Dialogue recommendation method and device, electronic equipment and storage medium

Abstract

The invention provides a dialogue recommendation method and device, electronic equipment and a storage medium. The method comprises the steps of obtaining dialogue data; performing intention reasoning on the dialogue data based on a large language model to obtain an intention reasoning thinking chain; performing retrieval based on the intention reasoning thinking chain to obtain recommended content; and performing dialogue recommendation based on the recommendation content. According to the method and device, the electronic equipment and the storage medium provided by the invention, intention reasoning is performed on the dialogue data through the large language model to obtain the intention reasoning thinking chain. Performing retrieval based on the intention reasoning thinking chain to obtain recommended content; and dialogue recommendation is performed based on the recommendation content, so that accurate, comprehensive and reliable intention reasoning is realized, the recommendation content fits the chat intention of the user, and smooth and natural dialogue recommendation is realized.

Description

Technical Field

[0001] The present invention relates to the field of human-computer interaction technology, and in particular to a dialogue recommendation method, device, electronic device and storage medium. Background Art

[0002] The traditional attribute-based dialogue recommendation solution uses a strategy module to achieve the most accurate recommendation in the shortest number of conversations. In practice, the relevant dialogue flow is designed, and the relevant slots are filled according to the user's response. For those that are not filled, a template is designed to ask questions until the necessary slots in the template are filled. Then the corresponding recommendation is made. Usually, the dialogue flow is implemented using technologies such as rule systems and reinforcement learning, and slot extraction is implemented using information extraction models.

[0003] However, the above-mentioned conversation recommendation method can only support relatively clear needs for recommendation. Once the needs are more inclined towards chatting, accurate conversation recommendations cannot be made because it is impossible to provide suggestions. Summary of the invention

[0004] The present invention provides a conversation recommendation method, device, electronic device and storage medium, which are used to solve the defect in the prior art that accurate conversation recommendation cannot be performed due to the inability to provide suggestions for the needs that are biased towards chatting.

[0005] The present invention provides a dialogue recommendation method, comprising: Get conversation data; Performing intention reasoning on the conversation data based on a large language model to obtain an intention reasoning thinking chain; Performing a search based on the intention reasoning chain to obtain recommended content; Based on the recommended content, a conversation recommendation is performed.

[0006] According to a dialogue recommendation method provided by the present invention, the intention reasoning of the dialogue data based on a large language model to obtain an intention reasoning thinking chain includes: Determine a target user profile of the user to which the conversation data belongs; Intention reasoning is performed on the conversation data and the target user portrait based on the large language model to obtain the intention reasoning thinking chain.

[0007] According to a conversation recommendation method provided by the present invention, the intention reasoning of the conversation data and the target user portrait based on the large language model to obtain the intention reasoning thinking chain includes: Based on the large language model, the conversation data is subjected to intention reasoning to obtain a basic intention reasoning thinking chain, and the basic intention reasoning thinking chain is subjected to intention reasoning in combination with the target user portrait to obtain the intention reasoning thinking chain.

[0008] According to a conversation recommendation method provided by the present invention, the target user portrait is obtained based on at least one of user identity information, preference information and user's historical recommendation content.

[0009] According to a conversation recommendation method provided by the present invention, the searching based on the intention reasoning thought chain to obtain the recommended content includes: Generate an intention recommendation instruction based on the reasoning result in the intention reasoning thinking chain; A search is performed based on the intention recommendation instruction to obtain the recommended content.

[0010] According to a method for recommending a conversation provided by the present invention, the conversation recommendation based on the recommended content includes: Obtaining overall response data based on the initial response data of the conversation data and the recommended content; Based on the overall reply data, dialogue recommendations are performed.

[0011] The present invention also provides a dialogue recommendation device, comprising: Get unit, get conversation data; An intention reasoning unit performs intention reasoning on the conversation data based on a large language model to obtain an intention reasoning thought chain; A recommendation unit, which performs retrieval based on the intention reasoning thought chain to obtain recommended content; The dialogue unit performs dialogue recommendation based on the recommended content.

[0012] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the conversation recommendation method described above is implemented.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described conversation recommendation methods.

[0014] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned dialogue recommendation methods.

[0015] The conversation recommendation method, device, electronic device and storage medium provided by the present invention perform intention inference on conversation data through a large language model to obtain an intention inference thought chain. Retrieval is performed based on the intention inference thought chain to obtain recommended content; based on the recommended content, conversation recommendation is performed, achieving accurate, comprehensive and reliable intention inference, so that the recommended content fits the user's chat intention, and achieving smooth and natural conversation recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 It is a structural diagram of a dialogue recommendation system in the prior art; Figure 2 This is one of the flowcharts of the conversation recommendation method provided by the present invention; Figure 3 This is the second flow chart of the conversation recommendation method provided by the present invention; Figure 4 It is a schematic diagram of the framework of the dialogue recommendation system provided by the present invention; Figure 5 is a structural schematic diagram of a dialogue recommendation device provided by the present invention; Figure 6 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Figure 1 It is a structural diagram of the dialogue recommendation system in the prior art, such as Figure 1 As shown in the figure, the dialogue recommendation system includes a dialogue module, a strategy module and a recommendation module. The user's input dialogue is input as the input query into the dialogue recommendation system, and the dialogue recommendation system responds to the system and makes dialogue recommendations. Among them, the dialogue module is mainly responsible for understanding the user's natural language input and giving corresponding feedback; the strategy module controls the dialogue process and determines whether it is chatting, asking questions, or making recommendations; the recommendation module is responsible for giving recommendation results based on the preferences provided by the user.

[0020] Based on the above-mentioned conversation recommendation system, the traditional implementation scheme in the existing technical scheme is attribute-based conversation recommendation. That is, the strategy module is used to achieve the most accurate recommendation in the shortest number of conversations. The relevant conversation flow is designed and the relevant slots are filled according to the user's reply. For those that are not filled, a template will be designed to ask questions until the necessary slots in the template are filled, and then the corresponding recommendation will be made. Usually, the implementation of the conversation flow will be completed by using technologies such as rule systems and reinforcement learning, and slot extraction is implemented using information extraction models.

[0021] However, the traditional slot solution can only support relatively clear needs for recommendation. Once the needs are more inclined to chat, slots cannot be provided. Taking movie and TV recommendations as an example, the traditional solution supports clear needs such as "recommend Xiao Ming's science fiction movies" very well. However, for chat needs such as "I want to watch a movie about Xiao Ming planting potatoes on Mars", it is difficult to extract slots, and thus it is difficult to make accurate conversation recommendations.

[0022] In view of the above problems, the present invention provides a conversation recommendation method to achieve accurate and comprehensive conversation recommendation in chat-type conversations. Figure 2 is one of the flow charts of the conversation recommendation method provided by the present invention, such as Figure 2 As shown, the method includes: Step 210, obtaining conversation data; Specifically, in the process of dialogue with the user, the text and voice data in the dialogue interaction process can be used as dialogue data. Among them, for the voice data input by the user, the voice data can be transcribed, and the transcribed text data can be used as dialogue data. The dialogue data here can be, for example, "I want to watch a movie about Xiao Ming planting potatoes on Mars." It should be noted that the dialogue data here can be the interaction data in the process of chatting, so the sentence structure in the dialogue data here is not uniform and fixed. Therefore, the user intention reflected in the dialogue data cannot be directly obtained through slot extraction.

[0023] Step 220, performing intention inference on the dialogue data based on a large language model to obtain an intention inference thinking chain; Here, the intention reasoning chain may include reasoning reasons and reasoning results. The reasoning reasons may be used to reflect the basis for intention reasoning of the dialogue data through the large language model; the reasoning results may be used to reflect the chat intention of the dialogue data. In addition, the large language model here may be, for example, the Spark large model.

[0024] Specifically, the conversation data can be input into a large language model, and the large language model can be used to perform intention reasoning on the conversation data, so as to achieve an in-depth understanding of the conversation data and obtain an intention recommendation thought chain. For example, the conversation data and the pre-built conversation recommendation prompt text can be input into the large language model, and the conversation recommendation prompt text can be used to instruct the large language model to perform intention reasoning on the conversation data, so as to obtain an intention reasoning thought chain. The conversation recommendation prompt text here can be "You are a conversation recommendation system, which satisfies the user's recommendation demands through conversation data, and can follow a specific format to generate an intention reasoning thought chain." For example, the user's conversation data is "I want to watch a movie in which Xiao Ming grows potatoes on Mars", and the intention reasoning thought chain can be: Reasoning basis: "Xiao Ming is the name of the actor, so the user may be interested in the film and television works starring this actor; Mars grows potatoes is the plot of the movie, so the user may be interested in movies involving Mars or potatoes"; Reasoning result: "The user wants to watch the movie "Movie A" starring Xiao Ming".

[0025] It is understandable that by performing intention reasoning on conversation data based on a large language model and outputting an intention reasoning chain that includes reasoning analysis basis and reasoning results, the accuracy, comprehensiveness and reliability of intention recognition on conversation data are improved with the help of the reasoning ability of a large language model. In particular, the accuracy of intention recognition in cases where user intentions are implicit in chat conversation data is improved, thereby enabling accurate, comprehensive and reliable intention reasoning, thereby improving the fluency of recommendations in chat conversations and achieving integrated conversation recommendations.

[0026] Step 230, searching based on the intention reasoning thought chain to obtain recommended content; The recommended content here can reflect the conversation topics that the user is interested in when outputting the conversation data.

[0027] Specifically, in the pre-built knowledge base, the inference results in the intention inference thinking chain can be used as search keywords to retrieve text content related to the inference results, and the related text content can be used as recommended content. For example, "the movie "Movie A" starring Xiao Ming" can be used as a search keyword to retrieve content related to this search keyword in the pre-built knowledge base to obtain recommended content. For example: the movie's introduction, play link, cast list and other information.

[0028] It should be noted that when the intention reasoning chain is empty, it means that the user's current conversation does not contain recommendation intentions, so no conversation recommendation can be made, and only the user's conversation data can be replied. Therefore, there is no need to configure templates or specially train related models to judge the conversation type, and the large language model as a whole can play the role of task planning.

[0029] Step 240: Perform a conversation recommendation based on the recommended content.

[0030] Specifically, the recommended content can be polished and the polished recommended content can be output to achieve a conversation recommendation with the user. Alternatively, the reply data of the conversation data and the recommended content can be spliced ​​and the spliced ​​data can be output to achieve a conversation recommendation with the user.

[0031] The method provided by the embodiment of the present invention uses a large language model to perform intention reasoning on the conversation data to obtain an intention reasoning thought chain. Based on the intention reasoning thought chain, retrieval is performed to obtain recommended content; based on the recommended content, conversation recommendations are performed, achieving accurate, comprehensive, and reliable intention reasoning, so that the recommended content fits the user's chat intention, and achieving smooth and natural conversation recommendations.

[0032] Based on any of the above embodiments, step 220 includes: Determine a target user profile of the user to which the conversation data belongs; Intention reasoning is performed on the conversation data and the target user portrait based on the large language model to obtain the intention reasoning thinking chain.

[0033] Here, the target user portrait can be information used to reflect the user identity information, hobbies, behavior patterns, psychological characteristics, etc. of the user corresponding to the conversation data. The target user portrait can provide a reasoning basis for intention reasoning in addition to the conversation data, which can make the intention reasoning thinking chain obtained by reasoning more accurate, comprehensive and reliable.

[0034] Specifically, a user portrait resource library can be constructed, which can contain user portraits of multiple users. For a single user portrait, the user portrait can be constructed by obtaining user information and the user's historical conversation data. After obtaining the conversation data, the user portrait of the currently logged-in account can be used as the target user portrait to which the conversation data belongs.

[0035] Then, the conversation data and the target user portrait can be input into a large language model, and the large language model can be used to combine the conversation data and the target user portrait to perform intention reasoning. For example, the conversation data can be deeply understood and further intention reasoning can be performed in combination with the target user portrait to obtain the final intention reasoning chain combined with the user portrait. For example, the conversation data is "The movie about growing potatoes on Mars is good", and the user portrait shows that the user is a loyal fan of science fiction movies and pays special attention to Xiao Ming's works. Therefore, by combining the conversation data and the user portrait for intention reasoning, the inference basis in the intention reasoning chain can be: the movie plot mentioned in the conversation data is "growing potatoes on Mars", and the target user portrait shows that "a loyal fan of science fiction movies and pays special attention to Xiao Ming's works". The inference result can be: the user wants to know about the movie "Movie A" starring Xiao Ming.

[0036] It should be noted that the existing technology is mainly the simple application of the pre-large model period and the large model. The main problem of the model in the pre-large model period is the coverage. It cannot cover the user's chat topic well, and will give a blunt reply such as "I don't know" or "the question is not related to this system". The construction of relevant dialogue flows requires a lot of manpower and a large amount of data stacking to cover relevant situations. Once the user needs change, it needs to be redeveloped. In addition, it is difficult to add incremental intents. The method provided in the embodiment of the present invention performs intent reasoning on dialogue data and target user portraits based on a large language model, obtains an intent reasoning thinking chain, and realizes deeper and more comprehensive intent reasoning on dialogue data, thereby realizing more comprehensive, accurate and reliable dialogue recommendations.

[0037] Based on any of the above embodiments, performing intention reasoning on the conversation data and the target user portrait based on the large language model to obtain the intention reasoning thinking chain includes: Based on the large language model, the conversation data is subjected to intention reasoning to obtain a basic intention reasoning thinking chain, and the basic intention reasoning thinking chain is subjected to intention reasoning in combination with the target user portrait to obtain the intention reasoning thinking chain.

[0038] Specifically, the conversation data and pre-built intent recommendation prompt text can be input into a large language model, and the intent recommendation prompt text can be used to instruct the large language model to analyze the conversation data to obtain a basic intent reasoning thinking chain based on the conversation data itself.

[0039] Furthermore, the target user portrait, basic intent reasoning thinking chain, and pre-built intent recommendation prompt text can be input into a large language model. The large language model can be combined with the target user portrait to perform further intent reasoning on the basic intent reasoning thinking chain to obtain a more comprehensive and accurate intent reasoning thinking chain.

[0040] It should be noted that here, the basic intention thinking chain is further analyzed in combination with the target user portrait. The basic intention thinking chain can be expanded through the information reflected by the target user portrait to obtain an incremental intention thinking chain related to the basic intention thinking chain. Alternatively, the basic intention thinking chain can be further determined through the information reflected by the target user portrait to obtain a clearer chat intention.

[0041] The method provided in the embodiment of the present invention performs intention inference on the dialogue data based on a large language model to obtain a basic intention inference thinking chain, and performs intention inference on the basic intention inference thinking chain in combination with the target user portrait to obtain a more accurate and closer to the user's real intention. In addition, it can realize incremental intentions related to the user, and then realize personalized dialogue recommendations for the user, greatly improving the interactive experience of dialogue recommendations.

[0042] Based on any of the above embodiments, the target user portrait is obtained based on at least one of user identity information, preference information and user's historical recommended content.

[0043] Specifically, the target user portrait can be constructed by obtaining at least one of the user identity information, preference information, and the user's historical recommended content. Among them, the user identity information is obtained, such as gender, age, occupation, marital status, family members, geographical location and other information. It can be understood that the user identity information is the basis for building a user portrait and helps to understand the user's basic background and characteristics. The user's preference information is obtained, such as interests, hobbies, values ​​and other information. For example, whether you like to watch domestic films or foreign films, and your favorite actors, etc.

[0044] In addition, user portraits are constructed through historical recommended content. For example, historical recommended content that has been recommended to users more than a preset threshold can be used to construct user portraits. For example, if the movie "Movie A" starring Xiao Ming was recommended to the user in the historical conversation, then the user's preference for watching the movie "Movie A" starring Xiao Ming can be used as the preference information in the user portrait. It can be understood that the historical recommended content here refers to the content recommended to the user in the previous conversation with the user, which can indicate that the user is interested in the recommended content, and then the target user portrait can be continuously enriched and improved based on the historical recommended content, so that subsequent conversation recommendations are closer to the user's actual intentions.

[0045] Based on any of the above embodiments, step 230 includes: Generate an intention recommendation instruction based on the reasoning result in the intention reasoning thinking chain; A search is performed based on the intention recommendation instruction to obtain the recommended content.

[0046] Specifically, first, the corresponding intention recommendation instruction can be generated through the reasoning results in the intention reasoning thinking chain. For example, if the reasoning result is "the user wants to know about the movie "Movie A" starring Xiao Ming", then the intention recommendation instruction can be "please search for content related to the movie "Movie A" starring Xiao Ming". It should be noted that the intention recommendation instruction is generated through the reasoning results in the intention reasoning thinking chain, and the reasoning results are obtained based on the reasoning reasons in the intention reasoning thinking chain. As a result, the intention recommendation instruction generated here is more accurate, and the recommended content obtained is more accurate.

[0047] Then, the intention recommendation instruction can be used to search in the pre-built external knowledge base. For example, a rough search is performed on the keywords in the intention recommendation instruction to obtain roughly recommended content. It can be understood that the roughly recommended content obtained here refers to the content related to the keywords. For another example, semantic matching can be performed on the inference results in the intention recommendation instruction, and the content that matches the semantics of the inference results can be matched from the external knowledge base to achieve a fine search and obtain fine recommended content. In this way, the roughly recommended content and / or the fine recommended content can be used as the final recommended content. It should be noted that the external knowledge base here can be a vector retrieval model or a recommendation model in other traditional recommendation systems, and the present invention does not specifically limit this.

[0048] Based on any of the above embodiments, step 240 includes: Obtaining overall response data based on the initial response data of the conversation data and the recommended content; Based on the overall reply data, dialogue recommendations are performed.

[0049] Specifically, first, the conversation data can be replied to through a large language model to obtain initial reply data of the conversation data. It should be noted that the initial reply data here is the data of direct conversation reply based on the user's conversation data to meet the user's basic needs or guide the user to further communicate.

[0050] Then, the obtained recommendation content can be inserted into the initial reply data to obtain the overall reply data. Finally, the overall reply data can be output to realize the conversation recommendation with the user.

[0051] The method provided by the embodiment of the present invention obtains overall reply data based on the initial reply data of the conversation data and the recommended content. Based on the overall reply data, the conversation recommendation is performed, and the interactive integrated conversation recommendation that fits the user's interaction intention and preference is realized on the basis of realizing the basic reply to the user's conversation data, without the need for the user's secondary interaction, thereby improving the interaction fluency and interaction efficiency, and thus improving the user's interaction experience.

[0052] Based on any of the above embodiments, Figure 3 FIG. 2 is a flow chart of the conversation recommendation method provided by the present invention. Figure 3 As shown, the method includes: First, the input content can be obtained, including the recommendation system prompt words, user portraits, and conversation history. Among them, the recommendation system prompt words are used to guide the large language model to perform intent reasoning and realize conversation process control. The recommendation system prompt words can be "As a conversation recommendation system, you communicate with users through conversations and meet the user's recommendation demands in combination with user portraits. You can use external recommendation system tools to find the most suitable recommended content. The output follows a fixed format. First, the user intention judgment logic chain is output. According to the intention judgment logic chain, specific chat or recommendation instructions are generated. If it is a recommendation, the recommendation system retrieval conditions are generated. After being sent to the recommendation system, a reply is generated after splicing the recommendation system return results. If the instruction is a chat, a reply is directly generated. "The user portrait contains basic information, preference information, and historical information of the user, that is, historical recommended content. The conversation history can contain multiple rounds of conversations between the user and the system, that is, conversation data. In addition, the recommended content does not exist in the conversation history.

[0053] Therefore, when the input content is input into the large model dialogue control module, the large model dialogue control module here can be constructed based on a large language model. The output content is obtained through the large model dialogue control module. The output content here includes instruction generation, recommended content and reply content. Among them, the instruction generation content is not output to the user, but is used to input the recommended instruction into the recommendation system when the specific instruction is a recommendation, and retrieve the relevant recommended content from the recommendation system. The recommended content here can include multiple.

[0054] It should be noted that when the large model dialogue control module generates instructions here, it is necessary to first output the user intention judgment logic chain, and then generate specific chat or recommendation instructions based on the intention judgment logic chain. For example, when the intention judgment logic chain is empty, the instruction type is a chat instruction. When the intention judgment logic chain is not empty, the instruction type is a recommendation instruction. In addition, the intention judgment logic chain contains the generation reason of the instruction, that is, the reasoning reason for obtaining the reasoning result.

[0055] Therefore, when chatting, the reply content is directly generated based on the user's conversation data. When recommending, the reply will be based on the recommended content inserted in the reply content, realizing conversation recommendation.

[0056] The method provided by the embodiment of the present invention uses a large model dialogue control system to grasp the dialogue process and decide whether to chat or make relevant recommendations. There is no need to configure a template or specially train relevant models for judgment. The large model as a whole can play the role of task planning. The prompt word project combines user portraits and interaction history to determine when to initiate a recommendation signal. The recommendation system uses a vector retrieval model and external additional information to retrieve relevant recommendation content candidates, and then returns to the large model for further sorting and regularization before making recommendations, achieving smooth, integrated dialogue recommendations that are close to user interaction intentions and preferences.

[0057] Based on any of the above embodiments, Figure 4 is a schematic diagram of the framework of the dialogue recommendation system provided by the present invention, such as Figure 4 As shown, the dialogue system includes a large model dialogue control system, a recommendation system, and a recommended content library. In the process of dialogue with the dialogue system, the large model dialogue recommendation system is used to grasp the dialogue process by combining the dialogue data input by the user and the user portrait, and decide whether to chat or make relevant recommendations. When it is determined that a recommendation is needed, a recommendation instruction is generated. The recommendation system retrieves the recommended content candidate in the recommended content library based on the recommendation instruction. Then, the recommendation system returns the recommended content candidate to the large model dialogue control system for insertion into the reply data of the dialogue data to obtain the final dialogue recommendation data to complete the dialogue recommendation with the user.

[0058] The dialogue recommendation system provided in the embodiment of the present invention constructs an overall systematic solution that makes use of large model capabilities, recommendation systems, and recommendation knowledge bases, uses user dialogue information and portrait information, and also combines external information to comprehensively analyze, understand, reason, and judge the content that users want, and gives the most suitable recommendations for the current situation, ultimately realizing a recommendation system under natural human-computer interaction, free interaction, and personalized interaction.

[0059] Based on any of the above embodiments, Figure 5 is a structural diagram of the dialogue recommendation device provided by the present invention, such as Figure 5 As shown, the device comprises: An acquisition unit 510 acquires conversation data; An intention reasoning unit 520 performs intention reasoning on the dialogue data based on a large language model to obtain an intention reasoning thought chain; A recommendation unit 530 performs a search based on the intention reasoning thought chain to obtain recommended content; The dialogue unit 540 performs dialogue recommendation based on the recommended content.

[0060] The device provided by the embodiment of the present invention performs intention reasoning on the conversation data through a large language model to obtain an intention reasoning thought chain. Based on the intention reasoning thought chain, retrieval is performed to obtain recommended content; based on the recommended content, conversation recommendations are performed, achieving accurate, comprehensive and reliable intention reasoning, so that the recommended content fits the user's chat intention, and achieving smooth and natural conversation recommendations.

[0061] Based on any of the above embodiments, the intention inference unit is specifically used for: Determine a target user profile of the user to which the conversation data belongs; Intention reasoning is performed on the conversation data and the target user portrait based on the large language model to obtain the intention reasoning thinking chain.

[0062] Based on any of the foregoing embodiments, the intention inference unit is further specifically used for: Based on the large language model, the conversation data is subjected to intention reasoning to obtain a basic intention reasoning thinking chain, and the basic intention reasoning thinking chain is subjected to intention reasoning in combination with the target user portrait to obtain the intention reasoning thinking chain.

[0063] Based on any of the above embodiments, the target user portrait is obtained based on at least one of user identity information, preference information and user's historical recommended content.

[0064] Based on any of the above embodiments, the recommendation unit is specifically used for: Generate an intention recommendation instruction based on the reasoning result in the intention reasoning thinking chain; A search is performed based on the intention recommendation instruction to obtain the recommended content.

[0065] Based on any of the above embodiments, the dialogue unit is specifically used for: Obtaining overall response data based on the initial response data of the conversation data and the recommended content; Based on the overall reply data, dialogue recommendations are performed.

[0066] Figure 6 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 6As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630 and a communication bus 640, wherein the processor 610, the communication interface 620 and the memory 630 communicate with each other through the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute the dialogue recommendation method, which includes: obtaining dialogue data; performing intention reasoning on the dialogue data based on a large language model to obtain an intention reasoning thinking chain; performing retrieval based on the intention reasoning thinking chain to obtain recommended content; and performing dialogue recommendation based on the recommended content.

[0067] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0068] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the dialogue recommendation method provided by the above methods, which method includes: obtaining dialogue data; performing intent reasoning on the dialogue data based on a large language model to obtain an intent reasoning thinking chain; performing retrieval based on the intent reasoning thinking chain to obtain recommended content; and performing dialogue recommendation based on the recommended content.

[0069] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it is implemented to execute the dialogue recommendation method provided by the above-mentioned methods, the method comprising: obtaining dialogue data; performing intention reasoning on the dialogue data based on a large language model to obtain an intention reasoning thinking chain; performing retrieval based on the intention reasoning thinking chain to obtain recommended content; and performing dialogue recommendation based on the recommended content.

[0070] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0071] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A conversation recommendation method, characterized in that: include: Get conversation data; Performing intention reasoning on the conversation data based on a large language model to obtain an intention reasoning thinking chain; Performing a search based on the intention reasoning chain to obtain recommended content; Based on the recommended content, a conversation recommendation is performed.

2. The method for recommending a conversation according to claim 1, characterized in that: The performing intention reasoning on the dialogue data based on the large language model to obtain an intention reasoning thinking chain includes: Determine a target user profile of the user to which the conversation data belongs; Intention reasoning is performed on the conversation data and the target user portrait based on the large language model to obtain the intention reasoning thinking chain.

3. The method for recommending a conversation according to claim 2, wherein: The performing intention reasoning on the conversation data and the target user portrait based on the large language model to obtain the intention reasoning thinking chain includes: Based on the large language model, the conversation data is subjected to intention reasoning to obtain a basic intention reasoning thinking chain, and the basic intention reasoning thinking chain is subjected to intention reasoning in combination with the target user portrait to obtain the intention reasoning thinking chain.

4. The method for recommending a conversation according to claim 2, wherein: The target user portrait is obtained based on at least one of user identity information, preference information and user's historical recommended content.

5. The method for recommending conversations according to any one of claims 1 to 4, characterized in that: The searching based on the intention reasoning thinking chain to obtain recommended content includes: Generate an intention recommendation instruction based on the reasoning result in the intention reasoning thinking chain; A search is performed based on the intention recommendation instruction to obtain the recommended content.

6. The method for recommending a conversation according to any one of claims 1 to 4, characterized in that: The performing a dialogue recommendation based on the recommended content includes: Obtaining overall response data based on the initial response data of the conversation data and the recommended content; Based on the overall reply data, dialogue recommendations are performed.

7. A conversation recommendation device, characterized in that: include: Get unit, get conversation data; An intention reasoning unit performs intention reasoning on the conversation data based on a large language model to obtain an intention reasoning thought chain; A recommendation unit, which performs retrieval based on the intention reasoning thought chain to obtain recommended content; The dialogue unit performs dialogue recommendation based on the recommended content.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the dialogue recommendation method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the conversation recommendation method according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the conversation recommendation method according to any one of claims 1 to 6 is implemented.

Images