Full-automatic multi-round business data query method and system

Abstract

The invention belongs to the technical field of business data query, and provides a full-automatic multi-round business data query method and system, and the method comprises the steps: obtaining newest query information, determining whether the newest query information comprises negative logic, if yes, reserving the last slot position information of the same type, and if not, directly entering the next step; and acquiring historical query information, comparing intentions and slot information of the latest query information and the historical query information, if any one of the intentions and the slot information of the latest query information and the historical query information is the same, performing multiple rounds of dialogues, otherwise, starting a new dialogue. According to the invention, a novel multi-round dialogue intention recognition logic is constructed, the accuracy and interaction efficiency of the system are improved, the working process of a multi-round dialogue mechanism is optimized, full-process automatic processing is realized, and the processing efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of business data query, and particularly relates to a fully automatic multi-round business data query method and system. Background Art

[0002] The statements in this part only provide background technical information related to the present invention, and do not necessarily constitute prior art.

[0003] When a user uses an artificial intelligence human-computer interaction system (hereinafter simply referred to as the AI system) for human-computer interaction dialogue, it is necessary to input a complete and cumbersome question each time when asking a question. Otherwise, the AI system cannot correctly understand the user's intention, reducing the integrity and accuracy of the answer.

[0004] Some existing solutions implement the multi-round dialogue function through prompt word splicing, that is, splicing the historical content input by the user, the historical content answered by the AI system, and the latest content input by the user to form a new prompt word and send it to the AI system, and the AI system answers the new prompt word content again. Although this method can achieve "continuous dialogue", it also causes the content of each sent message to be very large, and as the number of dialogue rounds increases, the token number of the prompt word will increase geometrically, not only reducing the answer efficiency of the AI system, but also increasing the user's usage cost. Summary of the Invention

[0005] In order to solve the above problems, the present invention proposes a fully automatic multi-round business data query method and system. The present invention constructs a new multi-round dialogue intention recognition logic, improves the accuracy and interaction efficiency of the AI system, optimizes the workflow of the multi-round dialogue mechanism, realizes full-process automatic processing, and improves the processing efficiency.

[0006] According to some embodiments, the present invention adopts the following technical solutions: A fully automatic multi-round business data query method includes the following steps: Obtain the latest query information, and determine whether it contains a negative logic. If so, retain the information of the last slot of the same type; if not, directly proceed to the next step; Obtain the historical query information, compare the intention and slot information of the latest query information and the historical query information. If any one of the intention and slot information of the two is the same, conduct a multi-round dialogue; otherwise, start a new dialogue.

[0007] As an alternative implementation, the process of determining whether it contains a negative logic includes determining whether there is a negative word in the latest query information. If so, according to the position of the negative word in the query, retain the information of the last slot of the same type that appears after the negative word as the true slot information.

[0008] As an alternative embodiment, the process of comparing the latest query information with the historical query information for intent and slot information includes: separately extracting the intent of the latest query information and the historical query information, separately extracting the slot information of the latest query information and the historical query information, comparing the two intents, comparing the two slot information, and if the two intents are the same, or the two slot information are the same, or the two intents are the same and the two slot information are the same, then proceed with multi-turn conversations; otherwise, directly start a new conversation.

[0009] As an alternative embodiment, it further includes the following steps: after multi-turn conversations, retain the latest intent information and delete the historical intent information. If the slot type is missing in the latest query information, perform semantic conjunction matching. If a match is found, retain the latest slot information, retain the historical slot information of different types, and delete the historical slot information of the same type; otherwise, retain the latest slot information and delete the historical slot information.

[0010] As a further embodiment, the process of performing semantic conjunction matching includes: constructing a semantic conjunction library, which contains common semantic conjunctions, and using keyword matching to detect whether the new conversation contains the words in the library. If it does, it is considered that the user wants to continue the conversation while retaining some historical information, that is, retain the latest slot information and the historical slot information of different types, and delete the historical slot information of the same type; if it does not, start a new conversation, that is, retain the latest slot information and delete the historical slot information.

[0011] A fully automatic multi-turn business data query system includes: An information acquisition module, configured to acquire the latest query information and determine whether it contains a negative logic. If so, retain the last slot information of the same type; if not, directly call the comparison module; A comparison module, configured to acquire the historical query information, compare the intent and slot information of the latest query information and the historical query information. If any one of the intent and slot information of the two is the same, then proceed with multi-turn conversations; otherwise, start a new conversation.

[0012] As an alternative embodiment, the process by which the information acquisition module is configured to determine whether it contains a negative logic includes determining whether there is a negative word in the latest query information. If so, according to the position of the negative word in the query, retain the last slot information of the same type that appears after the negative word as the true slot information.

[0013] As an alternative embodiment, the process of the comparison module configured to compare the intent and slot information of the latest query information and historical query information includes: separately extracting the intents of the latest query information and historical query information, separately extracting the slot information of the latest query information and historical query information, comparing the two intents, comparing the two slot information, and if the two intents are the same, or the two slot information are the same, or the two intents are the same and the two slot information are the same, then conduct multi-round conversations; otherwise, directly start a new conversation.

[0014] As an alternative embodiment, the following modules are further included: A semantic matching module, configured to, after conducting multi-round conversations, retain the latest intent information, delete the historical intent information, if the slot type is missing in the latest query information, conduct semantic conjunction matching, and if a match is found, retain the latest slot information, retain the historical slot information of different types, delete the historical slot information of the same type; otherwise, retain the latest slot information and delete the historical slot information.

[0015] As a further embodiment, the process of the semantic matching module configured to conduct semantic conjunction matching includes: constructing a semantic conjunction library, which contains common semantic conjunctions, and adopting a keyword matching method to detect whether the new conversation contains the words in the library. If it does, it is considered that the user wants to continue the conversation while retaining some historical information, that is, retain the latest slot information and the historical slot information of different types, and delete the historical slot information of the same type; if it does not contain, start a new conversation, that is, retain the latest slot information and delete the historical slot information.

[0016] Compared with the prior art, the beneficial effects of the present invention are: The present invention improves the processing logic of the multi-round conversation algorithm and enhances the method accuracy; at the same time, it does not require the user to participate in logical judgment, achieving fully automated judgment, and improving the processing efficiency and automation level.

[0017] By simultaneously comparing the intent and slot information, the present invention can more accurately identify the user's conversation needs, solves the problem in the prior art solution of over-relying on intent matching and ignoring the context semantic relevance, and can significantly improve the accuracy of multi-round conversations.

[0018] When the present invention compares the latest information and historical information, it first determines whether the user's question contains negative logic such as "no", "not", "wrong", etc. If it does, it selects the information of the last same slot as the true slot information, which can avoid misidentifying the existing slot and improve the accuracy of slot identification.

[0019] The present invention constructs a semantic connection lexicon and uses keyword matching to detect whether the new conversation contains words in the lexicon. If it does, it is considered that the user wants to continue the conversation while retaining some historical information, that is, retaining the latest slot information and different types of historical slot information, and deleting the same type of historical slot information; if it does not contain, a new conversation is started, that is, retaining the latest slot information and deleting the historical slot information. It can avoid the user's participation in logical judgment and achieve 100% fully automated processing without asking the user for further confirmation, improving the operation efficiency of the model while ensuring high accuracy.

[0020] To make the above objects, features and advantages of the present invention more obvious and understandable, the following specific preferred embodiments are given and described in detail in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The schematic diagrams in the specification forming a part of the present invention are used to provide a further understanding of the present invention. The schematic illustrative embodiments and their descriptions of the present invention are used to explain the present invention and do not constitute an improper limitation to the present invention.

[0022] Figure 1 is a schematic flow chart of an existing technical solution; Figure 2 is a schematic flow chart of a fully automated multi-round business data query method according to an embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] The present invention will be further described below in conjunction with the drawings and embodiments.

[0024] It should be noted that the following detailed descriptions are all illustrative and are intended to provide further explanations of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0025] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0026] In the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0027] Embodiment 1 As described in the background art, the solution for implementing the multi-round dialogue function through prompt splicing can achieve continuous dialogue, but there are certain problems in terms of efficiency and cost.

[0028] Some existing technical solutions attempt to solve the above problems, such as Figure 1 the multi-round dialogue method shown, which although improves the efficiency of the AI system to a certain extent, still has three problems: First, some judgment logics are missing. In scenarios where the intents are different but the slot types are partially the same, the existing technical solutions overly rely on intent matching and ignore the context semantic relevance, resulting in the inability to correctly handle cross-intent query requests with shared slots.

[0029] Second, the handling of negative semantics is missing. The existing technical solutions do not establish a negative logic recognition mechanism and cannot effectively handle the corrective negative expressions (such as "no", "not", "forbidden", etc.) in the user's dialogue, resulting in the residue of redundant slot information and semantic conflicts.

[0030] Third, manual intervention is required. When the slot type is missing in the existing technical solutions, intent confirmation needs to be carried out through human-computer interaction, resulting in the interruption of the query process and response delay, reducing the query efficiency.

[0031] To solve the above problems, this embodiment provides a fully automatic multi-round business data query method, which optimizes the overall process, improves the processing logic of the multi-round dialogue algorithm, and increases the accuracy of the method; at the same time, it does not require users to participate in logical judgment, achieving 100% automatic judgment and improving the processing efficiency and automation level.

[0032] A fully automatic multi-round business data query method, as Figure 2 shown, includes the following steps: Obtain the latest query information, determine whether it contains negative logic. If so, retain the last slot information of the same type; if not, directly proceed to the next step; Obtain the historical query information, compare the intent and slot information of the latest query information and the historical query information. If any one of the intent and slot information of the two is the same, conduct a multi-round dialogue; otherwise, start a new dialogue.

[0033] To solve the problem of missing handling of negative semantics, the process of determining whether it contains negative logic in this embodiment includes determining whether there are negative words in the latest query information. If so, according to the position of the negative word in the query, retain the last slot information of the same type that appears after the negative word as the true slot information.

[0034] That is, when comparing the latest information with the historical information in this embodiment, it is first determined whether the user's question contains negative logic such as "no", "not", "wrong", etc. If it contains, the information of the last same slot is selected as the true slot information.

[0035] Taking a specific example to explain: Suppose the new conversation is "Check how many contracts there were for 3 million last year. No, it's 5 million."

[0036] According to the process of the existing technology solution as Figure 1 shown, the obtained slot information is "Time: last year", "Amount 1: 3 million yuan", "Amount 2: 5 million yuan", and "Type: contract".

[0037] However, according to the context, the user clarified using negative logic in the question. Through the logic designed in the present invention, since the user's question contains negative logic and it is located that there are two slot categories of the same type, namely "amount", only the information of the last slot, that is, "Amount 2: 5 million yuan", is retained. Therefore, the correct slot information should be "Time: last year", "Amount: 5 million yuan", and "Type: contract".

[0038] Moreover, the process of this embodiment can avoid misidentifying the slots that already exist.

[0039] Taking a specific example to illustrate, suppose the new conversation is "How many contracts were there from 3 million to 5 million last year", the correct slot information should be "Time: last year", "Amount 1: 3 million yuan", "Amount 2: 5 million yuan", and "Type: contract".

[0040] Since the question does not contain negative logic, the above logic will not misdelete the slot information that should be extracted.

[0041] Taking another example to illustrate, suppose the new conversation is "Please query the travel guide for Beijing below 20,000 in budget, oh, no, it's for Nanjing, wrong, it's for Tokyo".

[0042] According to the process of the existing technology solution (the existing technology solution in this embodiment is the solution as Figure 1 shown, the same below), the obtained slot information is "Amount: 20,000", "Location 1: Beijing", "Location 2: Nanjing", "Location 3: Tokyo", and "Type: travel".

[0043] However, according to the context, the user clarified using negative logic in the question. Through the logic designed in the present invention, since the user's question contains negative logic and it is located that there are two slot categories of the same type, namely "location", only the information of the last slot, that is, "Tokyo", is retained. Therefore, the correct slot information should be "Amount: 20,000", "Location: Tokyo", and "Type: Travel".

[0044] On the other hand, to solve the problem of missing part of the judgment logic, the process of comparing the intent and slot information of the latest query information and the historical query information in this embodiment includes: separately extracting the intents of the latest query information and the historical query information, separately extracting the slot information of the latest query information and the historical query information, comparing the two intents, comparing the two slot information, and if the two intents are the same, or the two slot information are the same, or the two intents are the same and the two slot information are the same, then conduct multi-round conversations, otherwise directly start a new conversation.

[0045] The existing technical solution proposed in this embodiment is as Figure 1 shown. When comparing the intent of the latest information with the historical information, in the case of different intents, it is directly determined to start a new conversation. However, if there is a situation where the intents are different but the slot types are partially the same, correct classification cannot be performed.

[0046] Illustrated with a specific example: Suppose the historical conversation record is "Check how many invoices not exceeding 1 million yuan were there last month?", and the new conversation is "What about the number of contracts worth 3 million yuan?".

[0047] In these two conversations, the intents are "invoice query" and "contract query" respectively, and there is a slot of the same type, "amount".

[0048] According to the classification logic of the existing technical solution, it belongs to the "different intents" branch, so it is considered that the user wants to start a new conversation, and the obtained slot information is "Amount: 1 million yuan" and "Type: invoice".

[0049] However, from the context, the user's new conversation most likely wants to continue the query based on the previous conversation content, that is, the user's true intent is to query "How many contracts not exceeding 3 million yuan were there last month?", and the obtained slot information should be "Time: last month", "Amount: 3 million yuan", and "Type: contract".

[0050] Therefore, Figure 1 the existing technical solution shown over-relies on intent matching and ignores the context semantic relevance. The optimized algorithm improves the multi-round conversation accuracy rate from 94.3% to 98.1% on the test set.

[0051] Let's take another scenario as an example: Suppose that the historical conversation record is "Check how many express deliveries were sent in xx community in July" and the new conversation is "How many express deliveries were received?"

[0052] In these two conversations, the intents are "Query the quantity of express shipments" and "Query the quantity of express receipts", and there is a slot of the same type "Product: Express".

[0053] According to the classification logic of the existing technical solution, belonging to different branches is regarded as the user wanting to start a new conversation.

[0054] However, from the context, the user's new conversation is most likely to continue the query based on the content of the last conversation, that is, the user's real intention is to query "check how many express deliveries have been received in July?", and the slot information obtained should be "Time: July", "Location: xx Community" and "Product: Express Delivery", etc.

[0055] Finally, in order to solve the problem of requiring human intervention, this embodiment optimizes the judgment logic of the algorithm, avoids user participation in logical judgment while ensuring high accuracy, and achieves 100% fully automated processing.

[0056] Existing technical solutions, such as Figure 1 As shown in the figure, when the processing intent is the same but the slot type is missing, the user needs to be questioned and manually confirm the true intent. Although the high accuracy is maintained, the operating efficiency of the model will be greatly reduced.

[0057] In this embodiment, Figure 2 As shown, after multiple rounds of conversations, the latest intent information is retained and the historical intent information is deleted. If the slot type is missing in the latest query information, a semantic connective match is performed. If a match occurs, the latest slot information is retained, historical slot information of different types is retained, and historical slot information of the same type is deleted. Otherwise, the latest slot information is retained and the historical slot information is deleted.

[0058] The process of matching semantic connectives includes: building a semantic connective lexicon, wherein the semantic connective lexicon includes commonly used semantic connectives, such as "that", "this", "also", "what", "then", "again", etc., and using keyword matching to detect whether a new conversation contains words in the lexicon. If so, it is deemed that the user wants to continue the conversation while retaining some historical information, that is, retaining the latest slot information and different types of historical slot information, and deleting the same type of historical slot information; if not, a new conversation is started, that is, retaining the latest slot information and deleting the historical slot information.

[0059] Illustrated with a specific example, assume that the historical conversation record is "Check the number of invoices not exceeding 1 million yuan last month", and the new conversation is "Then what about checking the number of invoices this month?".

[0060] In the new conversation, there are two commonly used semantic connectors in Chinese, namely "then" and "ne", so it can be considered that the user wants to ask a new question based on the historical conversation. Therefore, the true meaning of the new conversation is "Check the number of invoices not exceeding 1 million yuan this month?", and the correct slot information should be "Time: This month", "Amount: 1 million yuan", and "Type: Invoice".

[0061] Assume that the historical conversation record is "Check the number of invoices not exceeding 1 million yuan last month", and the new conversation is "Check the number of invoices this month". Judging from the semantics alone, there is no semantic connector in the new conversation, indicating that there is no obvious correlation between the two questions. It can be considered that the user wants to reset the query information. Therefore, the correct slot information should be "Time: This month" and "Type: Invoice".

[0062] To highlight the advantages of the algorithm provided in this embodiment, a test set with the same intention and missing slot types is constructed. After testing, the accuracy rates of the present invention and the existing technical solutions are 98.5% and 98.7% respectively. It can be found that the above method does not significantly reduce the accuracy rate; however, by avoiding user participation, the average response time of the model is reduced from 3 seconds to 70 milliseconds. Therefore, the method provided in this embodiment greatly improves the running efficiency of the model and optimizes the user experience while maintaining a high accuracy rate.

[0063] Illustrated with another scenario example: Assume that the historical conversation record is "Check the travel guide for Beijing with a budget of 20,000 yuan in July", and the new conversation is "This, what if it is changed to October?".

[0064] In the new conversation, there are two commonly used semantic connectors in Chinese, namely "this" and "ne", so it can be considered that the user wants to ask a new question based on the historical conversation. Therefore, the true meaning of the new conversation is "Check the travel guide for Beijing with a budget of 20,000 yuan in October", and the correct slot information should be "Time: October", "Amount: 20,000 yuan", "Location: Beijing", and "Type: Travel guide".

[0065] For the steps not detailed in this embodiment, the relevant processes of the existing technical solutions can be referred to and will not be elaborated here.

[0066] Embodiment 2 A full-automatic multi-round business data query system, comprising: An information acquisition module, configured to acquire the latest query information, determine whether it contains a negative logic, and if so, retain the information of the last slot of the same type, and if not, directly call the comparison module; A comparison module, configured to acquire historical query information, compare the intent and slot information of the latest query information and the historical query information, and if any one of the intents and slot information of the two is the same, conduct a multi-round conversation, otherwise, start a new conversation.

[0067] As an alternative implementation, the process of the information acquisition module being configured to determine whether it contains a negative logic includes determining whether there is a negative word in the latest query information. If so, according to the position of the negative word in the query, retain the information of the last slot of the same type that appears after the negative word as the true slot information.

[0068] As an alternative implementation, the process of the comparison module being configured to compare the intent and slot information of the latest query information and the historical query information includes: respectively extracting the intents of the latest query information and the historical query information, respectively extracting the slot information of the latest query information and the historical query information, comparing the two intents, comparing the two slot information, and if the two intents are the same, or the two slot information are the same, or the two intents are the same and the two slot information are the same, then conduct a multi-round conversation, otherwise, directly start a new conversation.

[0069] As an alternative implementation, the following modules are further included: A semantic matching module, configured to retain the latest intent information and delete the historical intent information after a multi-round conversation. If the slot type is missing in the latest query information, perform a semantic conjunction matching. If a match is found, retain the latest slot information, retain the historical slot information of different types, and delete the historical slot information of the same type. Otherwise, retain the latest slot information and delete the historical slot information.

[0070] As a further implementation, the process of the semantic matching module being configured to perform a semantic conjunction matching includes: constructing a semantic conjunction library, the semantic conjunction library contains common semantic conjunctions, and in a keyword matching manner, detect whether the new conversation contains the words in the library. If so, it is considered that the user wants to continue the conversation on the premise of retaining some historical information, that is, retain the latest slot information and the historical slot information of different types, and delete the historical slot information of the same type; if not, start a new conversation, that is, retain the latest slot information and delete the historical slot information.

[0071] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD - ROM , optical memory, etc.).

[0072] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 block or multiple blocks.

[0073] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implement the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 block or multiple blocks.

[0074] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 block or multiple blocks.

[0075] The above is only the preferred embodiment of the present invention, and it is not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made by those skilled in the art without creative labor within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A fully automatic multi-round business data query method, characterized in that, Including the following steps: Obtain the latest query information, determine whether it contains negative logic. If so, retain the information of the last slot of the same type; if not, directly proceed to the next step. Obtain the historical query information, compare the intent and slot information of the latest query information and the historical query information. If any one of the intent and slot information of the two is the same, conduct a multi-round conversation; otherwise, start a new conversation.

2. The fully automatic multi-round service data query method according to claim 1, wherein The process of determining whether it contains negative logic includes determining whether there is a negative word in the latest query information. If so, according to the position of the negative word in the query, retain the information of the last slot of the same type that appears after the negative word as the true slot information.

3. The fully automatic multi-round service data query method according to claim 1, characterized in that, The process of comparing the intent and slot information of the latest query information and the historical query information includes: separately extracting the intent of the latest query information and the historical query information, separately extracting the slot information of the latest query information and the historical query information, comparing the two intents, comparing the two slot information. If the two intents are the same, or the two slot information are the same, or the two intents are the same and the two slot information are the same, then conduct a multi-round conversation; otherwise, directly start a new conversation.

4. A fully automatic multi-round business data query method according to any one of claims 1-3, characterized in that, It also includes the following steps: After conducting a multi-round conversation, retain the latest intent information, delete the historical intent information. If the slot type is missing in the latest query information, perform semantic conjunction matching. If a match is found, retain the latest slot information, retain the historical slot information of different types, and delete the historical slot information of the same type; otherwise, retain the latest slot information and delete the historical slot information.

5. The fully automatic multi-round service data query method according to claim 4, characterized in that The process of performing semantic conjunction matching includes: constructing a semantic conjunction library, which contains common semantic conjunctions. Adopt the keyword matching method to detect whether the new conversation contains the words in the library. If it contains, it is considered that the user wants to continue the conversation while retaining some historical information, that is, retain the latest slot information and the historical slot information of different types, and delete the historical slot information of the same type; if it does not contain, start a new conversation, that is, retain the latest slot information and delete the historical slot information.

6. A fully automatic multi-round business data query system, characterized in that Including: An information acquisition module, configured to obtain the latest query information, determine whether it contains negative logic. If so, retain the information of the last slot of the same type; if not, directly call the comparison module. A comparison module, configured to obtain the historical query information, compare the intent and slot information of the latest query information and the historical query information. If any one of the intent and slot information of the two is the same, conduct a multi-round conversation; otherwise, start a new conversation.

7. The fully automatic multi-round service data query system according to claim 6, characterized in that, The information acquisition module, configured to determine whether it contains negative logic, includes determining whether there is a negative word in the latest query information. If so, according to the position of the negative word in the query, retain the information of the last slot of the same type that appears after the negative word as the true slot information.

8. The fully automatic multi-round service data query system according to claim 6, characterized in that, The process of the comparison module configured to compare the intent and slot information of the latest query information and the historical query information includes: separately extracting the intents of the latest query information and the historical query information, separately extracting the slot information of the latest query information and the historical query information, comparing the two intents, comparing the two slot information, and if the two intents are the same, or the two slot information are the same, or the two intents are the same and the two slot information are the same, then conduct multi-round conversations; otherwise, directly start a new conversation.

9. A fully automatic multi-round service data query system according to any one of claims 6-8, characterized in that It further includes the following modules: A semantic matching module configured to, after conducting multi-round conversations, retain the latest intent information and delete the historical intent information. If the slot type is missing in the latest query information, conduct semantic conjunction matching. If a match is found, retain the latest slot information, retain the historical slot information of different types, and delete the historical slot information of the same type; otherwise, retain the latest slot information and delete the historical slot information.

10. A fully automatic multi-round service data query system as described in claim 9, characterized in that, The process of the semantic matching module configured to conduct semantic conjunction matching includes: constructing a semantic conjunction library that contains common semantic conjunctions, and using keyword matching to detect whether the new conversation contains the words in the library. If it does, it is considered that the user wants to continue the conversation while retaining some historical information, that is, retain the latest slot information and the historical slot information of different types, and delete the historical slot information of the same type; if it does not contain, start a new conversation, that is, retain the latest slot information and delete the historical slot information.

Images