Multi-round dialogue management system and method based on machine learning technology

Abstract

The invention belongs to the technical field of dialogue management. The invention provides a multi-round dialogue management system and method based on a machine learning technology, and the method comprises the steps: carrying out the prediction of the problem information through a first machine learning model in a process that a user inputs the problem information in a first dialogue column, and obtaining the dialogue intention of the user; according to the current dialogue intention, topic matching is carried out in dialogue historical information in a first dialogue column to obtain a plurality of matched historical dialogue information, and according to a set rule, simplification processing is carried out on each piece of historical dialogue information to obtain each piece of target historical dialogue information; according to the matching degree of each piece of target historical dialogue information and the current dialogue intention, determining a sorting sequence number, and according to the sorting sequence number, splicing each piece of target historical dialogue information and the current question information to obtain a second dialogue column. According to the invention, the overall dialogue quality and efficiency can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of dialogue management, and in particular to a multi-round dialogue management system and method based on machine learning technology. Background Art

[0002] Intelligent dialogue is the mainstream trend of current technological development. Through intelligent dialogue, users can be provided with logical and accurate answers to a variety of consulting questions, significantly improving the user's consulting efficiency and consulting experience. AI dialogue is mainly used in dialogue systems in e-commerce customer service, medical system customer service, or other application scenarios, such as chat rooms, study rooms, etc.

[0003] When users communicate with dialogue agents, discontinuities, i.e., multiple rounds of dialogue, often occur. For example, on the first day, a user communicates with an intelligent robot in the customer service system of a shopping app to inquire about the functions and size of product A. On the second and third days, the user inquires about the price discounts of other products. On the Nth day, the user inquires about the relevant information of product A again. However, since the user has made multiple inquiries, the dialogue bar at this time contains a lot of content that is not related to product A. If the user wants to view the consultation content of the first day, he must manually scroll up to find it, which makes it impossible for the user to conveniently refer to the consultation record of the first day to understand product A.

[0004] Therefore, how to splice related conversation content belonging to a certain topic in multiple rounds of conversations so that users can have a more complete and convenient understanding of the content of the topic is a technical problem that needs to be solved at present. Summary of the invention

[0005] In this regard, the present invention provides a multi-round dialogue management method, system, electronic device, computer storage medium and computer program product based on machine learning technology to solve the above-mentioned technical problems.

[0006] The present invention discloses a multi-round dialogue management method based on machine learning technology, which is applied to an intelligent body. The method comprises the following steps: in the process of a user inputting question information in a first dialogue bar, a first machine learning model is used to predict the question information to obtain the user's intention for this dialogue.

[0007] According to the present conversation intention, topic matching is performed in the conversation history information corresponding to the first conversation column to obtain a number of matching historical conversation information, and each of the historical conversation information is streamlined according to the set rules to obtain each target historical conversation information.

[0008] Determine a sorting sequence number according to the degree of match between each target historical conversation information and the intention of the current conversation, and splice each target historical conversation information with the current question information according to the sorting sequence number to obtain a second conversation bar; wherein the current question information in the second conversation bar is located on the homepage, and the target historical conversation information with a smaller sorting sequence number has a smaller distance from the current question information.

[0009] In some embodiments, the streamlining of each of the historical conversation information according to set rules to obtain each target historical conversation information includes: obtaining content input features in the process of the user inputting the current question information in the first conversation bar, the content input features at least including character input features and input content streamlining features, the character input features including character input rate and number of character modification times; obtaining the user's conversation intention obviousness level according to the content input features, and determining a simplification level according to the conversation intention obviousness level; wherein the simplification level is positively correlated with the conversation intention obviousness level; and streamlining each of the historical conversation information according to the simplification level to obtain each target historical conversation information.

[0010] In some embodiments, the step of deriving the user's conversation intention obvious level based on the content input feature analysis includes: using a second machine learning model to analyze the content input feature to derive the user's first conversation intention obvious level; retrieving several second conversation intention obvious levels associated with the user, and determining a fine-tuning value based on each of the second conversation intention obvious levels; wherein the second conversation intention obvious level is a conversation intention obvious level stored in a database that belongs to the same or related topic type as the current question information input by the user; and using the fine-tuning value to fine-tune the first conversation intention obvious level to obtain a second conversation intention obvious level.

[0011] In some embodiments, before determining the sorting sequence according to the degree of match between each target historical conversation information and the intention of the current conversation, it also includes: outputting a prompt message in the first conversation bar, wherein the prompt message is used to ask the user whether to turn on multi-round conversation integration processing; if a confirmation signal from the user for the prompt message is received, an integration processing trigger signal is generated, wherein the integration processing trigger signal is used to trigger the determination of the sorting sequence according to the degree of match between each target historical conversation information and the intention of the current conversation.

[0012] In some embodiments, after obtaining the second conversation bar, it also includes: having a conversation with the user in the second conversation bar, and after the conversation ends, adding the new conversation content in the second conversation bar to the first conversation bar according to the conversation time.

[0013] The present invention also discloses a multi-round dialogue management system based on machine learning technology, which is applied to an intelligent body. The system includes a processor, a memory, and a computer program that can be run on the processor is stored in the memory. When the computer program is run by the processor, the following steps are performed: in the process of a user inputting the current question information in a first dialogue bar, the current question information is predicted using a first machine learning model to obtain the user's current dialogue intention; topic matching is performed in the dialogue history information corresponding to the first dialogue bar according to the current dialogue intention to obtain a number of matching historical dialogue information, and each of the historical dialogue information is streamlined according to set rules to obtain each target historical dialogue information; a sorting number is determined according to the matching degree of each target historical dialogue information with the current dialogue intention, and each of the target historical dialogue information and the current question information are spliced ​​according to the sorting number to obtain a second dialogue bar; wherein the current question information in the second dialogue bar is located at the homepage position, and the target historical dialogue information with a smaller sorting number has a smaller distance from the current question information.

[0014] In some embodiments, the streamlining of each of the historical conversation information according to set rules to obtain each target historical conversation information includes: obtaining content input features in the process of the user inputting the current question information in the first conversation bar, the content input features at least including character input features and input content streamlining features, the character input features including character input rate and number of character modification times; obtaining the user's conversation intention obviousness level according to the content input features, and determining a simplification level according to the conversation intention obviousness level; wherein the simplification level is positively correlated with the conversation intention obviousness level; and streamlining each of the historical conversation information according to the simplification level to obtain each target historical conversation information.

[0015] In some embodiments, the step of deriving the user's conversation intention obvious level based on the content input feature analysis includes: using a second machine learning model to analyze the content input feature to derive the user's first conversation intention obvious level; retrieving several second conversation intention obvious levels associated with the user, and determining a fine-tuning value based on each of the second conversation intention obvious levels; wherein the second conversation intention obvious level is a conversation intention obvious level stored in a database that belongs to the same or related topic type as the current question information input by the user; and using the fine-tuning value to fine-tune the first conversation intention obvious level to obtain a second conversation intention obvious level.

[0016] In some embodiments, before determining the sorting sequence according to the degree of match between each target historical conversation information and the intention of the current conversation, it also includes: outputting a prompt message in the first conversation bar, wherein the prompt message is used to ask the user whether to turn on multi-round conversation integration processing; if a confirmation signal from the user for the prompt message is received, an integration processing trigger signal is generated, wherein the integration processing trigger signal is used to trigger the determination of the sorting sequence according to the degree of match between each target historical conversation information and the intention of the current conversation.

[0017] In some embodiments, after obtaining the second conversation bar, it also includes: having a conversation with the user in the second conversation bar, and after the conversation ends, adding the new conversation content in the second conversation bar to the first conversation bar according to the conversation time.

[0018] The present invention also discloses an electronic device, applied to an intelligent body, the electronic device comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program implements the method described in any of the preceding items when executed.

[0019] The present invention also discloses a computer storage medium, which is applied to an intelligent agent. The computer-readable storage medium stores a computer program, and when the computer program is executed, the method described in any of the preceding items is implemented.

[0020] The present invention also discloses a computer program product, which is applied to an intelligent body. The computer program product is pre-packaged with computer program codes, and when the computer program codes are executed, the method described in any of the preceding items is implemented.

[0021] The beneficial effects of the present invention are as follows: based on the predicted conversation intention, the present invention can quickly retrieve matching information from multiple rounds of conversation history, saving users time in searching for related past topics and improving conversation coherence. Moreover, key details in past conversations can also be quickly reused, avoiding users from repeating their elaborations and optimizing the interactive experience, thereby improving the overall conversation quality and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0023] Figure 1 It is a structural diagram of a multi-round dialogue management method based on machine learning technology disclosed in an embodiment of the present invention.

[0024] Figure 2 It is a structural diagram of a multi-round dialogue management system based on machine learning technology disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following is a description of the implementation of the present application by specific specific embodiments. People familiar with the technology can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0026] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0027] like Figure 1 As shown, an embodiment of the present invention discloses a multi-round dialogue management method based on machine learning technology, which is applied to a dialogue agent. The method includes the following steps: S1. During the process of the user inputting the question information in the first dialogue bar, the first machine learning model is used to predict the question information to obtain the user's intention for the dialogue.

[0028] In this step, the dialogue agent monitors the user's input operation in the first dialogue box and captures the input stream character by character or word by word. Once the user starts to enter the question information, the preprocessing process is immediately started.

[0029] First, remove special characters such as "#", "&", "*", etc., which do not carry key semantic information but will interfere with subsequent analysis. For the cleaned question information, split it into individual words or subword units.

[0030] Then, we use pre-trained machine learning models, such as a multi-layer bidirectional long short-term memory network (Bi-LSTM) combined with an attention mechanism (Attention) architecture. This type of model is good at capturing long-distance dependencies in text sequences and performs well in conversation intent classification tasks. We pre-load model parameters that have been fine-tuned on a large-scale conversation dataset that covers various daily, business, customer service and other conversation scenarios to ensure the generalization ability of the model.

[0031] The preprocessed text is segmented, and each word is mapped into a vector of fixed dimension using word embedding technology (such as GloVe) and input into the Bi-LSTM model. The model performs forward and backward traversal learning on the text sequence, and the attention mechanism focuses on the key semantic parts. The final output layer is passed through the softmax function to give different conversation intent categories. The category with the highest probability is the predicted intention of this conversation. For example, the intention of this conversation is "the type of motor used by A's washing machine", "how B technology is used for its chat intention analysis", etc.

[0032] S2, performing topic matching in the conversation history information corresponding to the first conversation column according to the current conversation intention, obtaining a number of matching historical conversation information, and simplifying each of the historical conversation information according to a set rule to obtain each target historical conversation information.

[0033] In this step, the predicted intention of this conversation is used as an index to query the conversation history information belonging to the first conversation column stored in the database. The database uses a relational database or a document database to store structured conversation records, and each record has fields such as the conversation column, timestamp, conversation role, and conversation content. The fuzzy query function of the SQL statement or the full-text search function of the document database is used to search for paragraphs related to this intention in past conversations, and a batch of matching historical conversation information is extracted. For example, if the intention of this time is "how B technology is used for its chat intention analysis", historical conversations containing keywords such as "B technology", "chat", and "intention" are searched. Among them, each matched historical conversation information is also associated with a corresponding matching degree, which is used to characterize the degree of semantic matching between the historical conversation information and the intention of this conversation.

[0034] At the same time, the matched historical conversation information may also contain a lot of irrelevant information, such as "Are you there?", "hello", overly long background description information, etc. These information irrelevant to the core topic need to be streamlined first to ensure that the final spliced ​​conversation page is more streamlined and efficient.

[0035] Simplification rules include: 1) removing repeated semantic fragments. If several consecutive sentences describe the same store address, only the most complete and clear sentence is retained; 2) filtering out meaningless modal particles and catchphrases, such as "um", "ah", and "that"; 3) trimming overly long background descriptions and only retaining key information directly related to the core topic. Natural language processing tools, such as part-of-speech tagging and dependency syntactic analysis, are used to assist in determining semantic relevance, complete the simplification of each historical dialogue information, and obtain the target historical dialogue information.

[0036] After the above simplification process, each matched historical conversation information is trimmed into the target historical conversation information. The matched historical conversation information refers to the historical conversation information with a matching degree higher than a preset matching threshold.

[0037] It should be noted that for the matching degree between each target historical conversation information and the intention of this conversation, a semantic similarity algorithm can be applied, such as cosine similarity based on semantic vector space. First, the intention of this conversation and the target historical conversation are vectorized separately, and then the cosine value of the angle between the two vectors is calculated. The closer the value is to 1, the higher the matching degree is. At the same time, combined with the vocabulary overlap rate, the proportion of the number of key nouns and verbs shared by the two is counted, and the matching degree score is comprehensively obtained. Then, the historical conversation information with a matching degree higher than the preset matching threshold is screened out.

[0038] S3, determining a sorting sequence number according to the matching degree between each target historical conversation information and the intention of the current conversation, and concatenating each target historical conversation information and the current question information according to the sorting sequence number to obtain a second conversation bar; wherein the current question information in the second conversation bar is located on the home page, and the target historical conversation information with a smaller sorting sequence number has a smaller distance from the current question information.

[0039] In this step, after obtaining the historical dialogue information of each target that is highly relevant to the current dialogue intention, the historical dialogue information of each target and the current question information can be spliced ​​to obtain the second dialogue column. In the second dialogue column, the historical rounds of dialogue that are irrelevant to the current dialogue intention are hidden, and only the current question information and the historical dialogue information of each target are retained, wherein the current question information is located at the home page of the page, that is, the "latest" position of the current dialogue page.

[0040] As for each target historical conversation information, its position in the second conversation column can be determined according to the degree of its match with the intention of this conversation. Specifically, for the target historical conversation information with a higher degree of match, the closer it is to the current question information, the closer it is to the above-mentioned "latest" position. In other words, it is placed in the front; and for the target historical conversation information with a lower degree of match, the farther it is from the current question information, the closer it is to the above-mentioned "latest" position. In other words, it is placed in the back.

[0041] Based on the predicted conversation intention, the present invention can quickly retrieve matching information from multiple rounds of conversation history, saving users time in searching for related past topics and improving conversation coherence. In addition, key details in past conversations can be quickly reused to avoid users from repeating their elaborations, optimize the interactive experience, and thus improve the overall conversation quality and efficiency.

[0042] In some embodiments, the streamlining of each of the historical conversation information according to set rules to obtain each target historical conversation information includes: obtaining content input features in the process of the user inputting the current question information in the first conversation bar, the content input features at least including character input features and input content streamlining features, the character input features including character input rate and number of character modification times; obtaining the user's conversation intention obviousness level according to the content input features, and determining a simplification level according to the conversation intention obviousness level; wherein the simplification level is positively correlated with the conversation intention obviousness level; and streamlining each of the historical conversation information according to the simplification level to obtain each target historical conversation information.

[0043] In the embodiment of the present invention, character input rate: the dialogue agent starts the timing module from the moment the user starts to input the question information in the first dialogue bar. The character input rate is obtained by counting the number of characters input by the user per unit time (e.g., per second). If the user quickly types 5 characters in 1 second, it means that the input rhythm is fast and the user may have a clear idea of ​​the question. On the contrary, if the user only types 1 character in 3 seconds, the user is likely still thinking about the core content of the question.

[0044] Number of character modifications: The system monitors the user's modification operations on the entered characters in real time, and every deletion and replacement of characters is recorded. A high number of modifications may indicate that the user is still organizing accurate expression and has not yet fully clarified his or her own demands.

[0045] The above-mentioned character input rate and character modification times constitute the character input features.

[0046] After the user has completed the final input, the user's input content is analyzed and refined. The character ratio of the extracted essential content and the entire input content is compared, and the character ratio is the input content simplification feature. Obviously, the higher the character ratio, the more concise the user's input content is, and vice versa, the more verbose it is (that is, it contains too many irrelevant words).

[0047] The character input features and input content simplification features extracted above constitute the content input features of the user, based on which the user's conversation intention level can be analyzed. The conversation intention level here refers to the degree of clarity of the core content of the question to be consulted or the related content involved when the user is inputting content. In other words, the higher the conversation intention level, the clearer the user knows the core content of the question to be consulted or the related content involved, and vice versa.

[0048] After determining the level of obviousness of the conversation intention, the level of simplification can be determined according to the positive correlation. The level of simplification here refers to the intensity of the simplification processing of each historical conversation information. The higher the simplification level, the higher the intensity of the simplification processing, and the lower the simplification level, the lower the intensity of the simplification processing. Simplification processing includes but is not limited to deletion and semantic integration. For example, the historical conversation information is: Agent: Hello, how can I help you? User: I am currently studying technology B and specifically analyzing its feasibility in chat intent analysis. Can you give me some guidance? Agent: Okay, technology B is…chat intent analysis is…

[0049] User: I know these background contents. Please tell me directly how to use B technology to analyze chat intent.

[0050] Agent: OK, … (specific suggestions on applying B technology to chat intent analysis).

[0051] For low-precision simplification processing, the above historical dialogue information can be simplified to: (Deleted: Agent: Hello, how can I help you?) User: I am currently studying technology B and specifically analyzing its feasibility in chat intent analysis. (Deleted: Can you give me relevant guidance?) Agent: (Deleted: OK,) Technology B is…Chat intent analysis is….

[0052] User: (Deleted: I know these background contents.) Please tell me directly how to use B technology for chat intent analysis.

[0053] Agent: (Deleted: OK,) …(Specific suggestions on applying B technology to chat intent analysis).

[0054] For high-precision simplification processing, the above historical conversation information can be simplified to: (Deleted: Agent: Hello, how can I help you?) (Deleted: User: I am currently studying technology B and specifically analyzing its feasibility in chat intent analysis. Can you give me some guidance?) (Deleted: Agent: OK, technology B is…chat intent analysis is….) User: (Deleted: I know these background contents.) Please tell me directly how to use B technology for chat intent analysis.

[0055] Agent: (Deleted: OK,) …(Specific suggestions on applying B technology to chat intent analysis).

[0056] The present invention analyzes the obvious level of the user's conversation intention to determine the degree of simplification of each historical conversation information, thereby realizing personalized simplification of the historical conversation information, thereby ensuring that the second conversation column constructed subsequently includes more effective information and does not omit key information. Specifically, when the user knows more clearly the core content of the question to be consulted or the related content involved, the historical conversation information is simplified with higher intensity, so that the second conversation column can be more streamlined; when the user knows less clearly the core content of the question to be consulted or the related content involved, the historical conversation information is simplified with lower intensity, so that the second conversation column can include more detailed content of the historical conversation information, which is conducive to providing users with more conversation references.

[0057] In some embodiments, the step of deriving the user's conversation intention obvious level based on the content input feature analysis includes: using a second machine learning model to analyze the content input feature to derive the user's first conversation intention obvious level; retrieving several second conversation intention obvious levels associated with the user, and determining a fine-tuning value based on each of the second conversation intention obvious levels; wherein the second conversation intention obvious level is a conversation intention obvious level stored in a database that belongs to the same or related topic type as the current question information input by the user; and using the fine-tuning value to fine-tune the first conversation intention obvious level to obtain a second conversation intention obvious level.

[0058] In an embodiment of the present invention, a machine learning algorithm suitable for processing small data volumes and high-dimensional features is selected to build the second machine learning model, such as a lightweight gradient boosting tree (LightGBM). A large amount of historical user input data is collected as a training set. These data cover a rich content input feature, such as a variety of character input rates, character modification times, and sample combinations of input content simplification features, and the corresponding level of dialogue intent is annotated. After multiple rounds of training, the model learns the potential mapping relationship between features and levels.

[0059] After obtaining the content input features of the user's input of the question information in the first dialogue bar, organize them into a format that meets the model input requirements, such as digitizing the character input rate and the number of character modifications into a vector form, and input them into the trained second machine learning model. The second machine learning model outputs a preliminary level of the user's first dialogue intention based on the learned potential mapping relationship. The level division can be qualitative descriptions such as "low", "medium", and "high", or specific level values, providing a basis for subsequent fine-tuning.

[0060] At the same time, the present invention also obtains the obvious level of the user's conversation intention for the same or similar type of topic stored in the database. These obvious levels of conversation intention are analyzed by the intelligent agent when determining the second conversation column of the present invention, that is, each time multiple rounds of conversation management are performed, the obvious level of conversation intention obtained by the analysis is connected to the topic type and stored in the database. The extracted several obvious levels of second conversation intention are statistically analyzed, for example, the mean of each obvious level of second conversation intention is obtained, and the fine-tuning value is determined according to the mean. If most of these past levels are "high", it means that the user has always had clear intentions on such topics, then the fine-tuning value is set to be closer to 1, reducing the interference with the obvious level of the first conversation intention this time; if most of the past levels are "low", it means that there is a probability that the confidence of the obvious level of the first conversation intention obtained by this analysis is "high", and the fine-tuning value is set to be less than 1. Among them, the association relationship between the fine-tuning value and the above-mentioned mean can be established in advance, and the corresponding fine-tuning value can be determined by querying the association relationship.

[0061] In some embodiments, before determining the sorting sequence according to the degree of match between each target historical conversation information and the intention of the current conversation, it also includes: outputting a prompt message in the first conversation bar, wherein the prompt message is used to ask the user whether to turn on multi-round conversation integration processing; if a confirmation signal from the user for the prompt message is received, an integration processing trigger signal is generated, wherein the integration processing trigger signal is used to trigger the determination of the sorting sequence according to the degree of match between each target historical conversation information and the intention of the current conversation.

[0062] In the embodiment of the present invention, in order to improve the user experience, the user can be consulted in advance whether he has the need to integrate the multi-round dialogue management. Only when it is confirmed that the user has this need, the multi-round dialogue management is performed and only the second dialogue bar is generated. When the second dialogue bar is not generated, the user can manually scroll up to find the historical dialogue content.

[0063] In some embodiments, after obtaining the second conversation bar, it also includes: having a conversation with the user in the second conversation bar, and after the conversation ends, adding the new conversation content in the second conversation bar to the first conversation bar according to the conversation time.

[0064] In an embodiment of the present invention, the first dialogue bar and the second dialogue bar involved in the present invention may be the same dialogue bar or different dialogue bars. For example, the first dialogue bar is a separate dialogue bar opened by a user with a certain merchant, a certain topic person / anchor, which contains multiple rounds of historical dialogues; and the second dialogue bar may also be the first dialogue bar, but the historical dialogues in the first dialogue bar are filtered and spliced ​​based on the intention of this dialogue. The second dialogue bar may also be a blank dialogue bar directly reopened by the intelligent body, and the content obtained by the above-mentioned filtering and splicing is transmitted to the blank dialogue bar. After the user ends this dialogue, the "blank dialogue bar" is deleted, and the content of this dialogue is stored in the "latest" position in the original first dialogue bar in chronological order.

[0065] like Figure 2 As shown, the present invention also discloses a multi-round dialogue management system based on machine learning technology, which is applied to an intelligent body. The system includes a processor, a memory, and a computer program that can be run on the processor is stored in the memory. When the computer program is run by the processor, the following steps are performed: a sorting sequence number is determined according to the matching degree between each target historical dialogue information and the intention of the current dialogue, and each target historical dialogue information and the current question information are spliced ​​according to the sorting sequence number to obtain a second dialogue bar; wherein the current question information in the second dialogue bar is located on the homepage, and the target historical dialogue information with a smaller sorting sequence number has a smaller distance from the current question information.

[0066] An embodiment of the present invention also discloses an electronic device, applied to an intelligent body, wherein the electronic device comprises: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program implements the method described in any of the preceding items when executed.

[0067] An embodiment of the present invention further discloses a computer storage medium, which is applied to an intelligent agent. The computer-readable storage medium stores a computer program, and when the computer program is executed, the method described in any of the preceding items is implemented.

[0068] An embodiment of the present invention further discloses a computer program product, which is applied to an intelligent agent. The computer program product is pre-packaged with computer program code, and when the computer program code is executed, the method described in any of the preceding items is implemented.

[0069] It should be noted that, in the description of the present invention, the terms "first", "second", etc. are used for descriptive purposes only and cannot be understood as indicating or implying relative importance. In addition, in the description of the present invention, unless otherwise specified, "multiple" means two or more. Any process or method description in the flowchart or otherwise described herein may be understood to represent a module, fragment or portion of a code including one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes additional implementations, in which functions may not be performed in the order shown or discussed, including in a substantially simultaneous manner or in reverse order according to the functions involved, which should be understood by a technician in the technical field to which the embodiments of the present invention belong.

[0070] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0071] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A multi-round dialogue management method based on machine learning technology, applied to an intelligent agent, characterized in that: The method comprises the following steps: when the user inputs the current question information in the first dialogue bar, using the first machine learning model to predict the current question information to obtain the user's current dialogue intention; According to the intention of this conversation, topic matching is performed in the conversation history information corresponding to the first conversation column to obtain a number of matching historical conversation information, and each of the historical conversation information is streamlined according to the set rules to obtain each target historical conversation information; a sorting sequence number is determined according to the matching degree of each target historical conversation information with the intention of this conversation, and each target historical conversation information and the current question information are spliced ​​according to the sorting sequence number to obtain a second conversation column; wherein the current question information in the second conversation column is located on the homepage, and the target historical conversation information with a smaller sorting sequence number has a smaller distance from the current question information.

2. The multi-round dialogue management method based on machine learning technology according to claim 1, characterized in that: Each of the historical conversation information is simplified according to the set rules to obtain each target historical conversation information, including: obtaining content input features in the process of the user inputting the current question information in the first conversation column, the content input features at least including character input features and input content simplification features, the character input features including character input rate and number of character modification times; obtaining the user's conversation intention obvious level according to the content input features, and determining the simplification level according to the conversation intention obvious level; wherein the simplification level is positively correlated with the conversation intention obvious level; and simplifying each of the historical conversation information according to the simplification level to obtain each target historical conversation information.

3. The multi-round dialogue management method based on machine learning technology according to claim 2, characterized in that: The method comprises: analyzing the content input features using a second machine learning model to analyze the content input features to obtain a first level of obvious conversation intention of the user; retrieving a plurality of second levels of obvious conversation intention associated with the user, and determining a fine-tuning value based on each of the second levels of obvious conversation intention; wherein the second level of obvious conversation intention is a level of obvious conversation intention stored in a database that belongs to the same or related topic type as the current question information input by the user; and fine-tuning the first level of obvious conversation intention using the fine-tuning value to obtain a second level of obvious conversation intention.

4. The multi-round dialogue management method based on machine learning technology according to claim 1, characterized in that: Before determining the sorting sequence according to the degree of match between each target historical conversation information and the intention of the current conversation, it also includes: outputting a prompt message in the first conversation bar, the prompt message is used to ask the user whether to start multi-round conversation integration processing; if a confirmation signal from the user for the prompt message is received, an integration processing trigger signal is generated, and the integration processing trigger signal is used to trigger the determination of the sorting sequence according to the degree of match between each target historical conversation information and the intention of the current conversation.

5. The multi-round dialogue management method based on machine learning technology according to claim 1, characterized in that: After obtaining the second dialogue bar, the method further includes: having a dialogue with the user in the second dialogue bar, and after the dialogue ends, adding the newly added dialogue content in the second dialogue bar to the first dialogue bar according to the dialogue time.

6. A multi-round dialogue management system based on machine learning technology, applied to an intelligent agent, characterized in that: The system includes a processor, a memory, and a computer program that can be run on the processor is stored in the memory. When the computer program is run by the processor, the following steps are performed: when the user inputs the current question information in the first dialogue bar, the first machine learning model is used to predict the current question information to obtain the user's current dialogue intention; According to the intention of this conversation, topic matching is performed in the conversation history information corresponding to the first conversation column to obtain a number of matching historical conversation information, and each of the historical conversation information is streamlined according to the set rules to obtain each target historical conversation information; a sorting sequence number is determined according to the matching degree of each target historical conversation information with the intention of this conversation, and each target historical conversation information and the current question information are spliced ​​according to the sorting sequence number to obtain a second conversation column; wherein the current question information in the second conversation column is located on the homepage, and the target historical conversation information with a smaller sorting sequence number has a smaller distance from the current question information.

7. The multi-round dialogue management system based on machine learning technology according to claim 6, characterized in that: The method of simplifying each of the historical conversation information according to the set rules to obtain each target historical conversation information includes: obtaining content input features in the process of the user inputting the current question information in the first conversation bar, the content input features at least including character input features and input content simplification features, the character input features including character input rate and number of character modification times; obtaining the user's conversation intention obviousness level according to the content input features, and determining the simplification level according to the conversation intention obviousness level; wherein the simplification level is positively correlated with the conversation intention obviousness level; and simplifying each of the historical conversation information according to the simplification level to obtain each target historical conversation information.

8. An electronic device, applied to an intelligent agent, characterized in that: The electronic device comprises: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program implements the method according to any one of claims 1 to 5 when executed.

9. A computer storage medium, applied to an intelligent agent, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed, the method according to any one of claims 1 to 5 is implemented.

10. A computer program product, applied to an intelligent agent, characterized in that: The computer program product is prepackaged with computer program codes, and when the computer program codes are executed, the method according to any one of claims 1 to 5 is implemented.

Images