A method and system for flexibly constructing multi-round dialogues in business scenarios

Abstract

The present invention discloses a method and system for flexibly constructing multi-round dialogues in business scenarios, including constructing a start node, a logic judgment node, an end node and a connecting line to construct a data flow; wherein the start node serves as the beginning of the entire process and supports the configuration of the common mandatory end corpus of the entire dialogue; the connecting line between the start node and the logic judgment node constructs the data flow diagram into a complex dialogue scenario; the logic judgment node provides an external service to provide functions to the outside world and supports connecting multiple complex dialogue scenarios; the end node is used to summarize information and return the information the user wants to the user after processing. The present invention extracts information through nodes, supports custom coding operations to complete scene design, solves the problem that the application scenarios of the prior art are relatively fixed and cannot realize flexible switching of scenes to meet the complex scene links; the use of connecting lines for subsequent logical judgment is clearer and can quickly and intuitively establish dialogue flows for various complex scenarios.

Description

Technical Field

[0001] The present invention relates to the field of intelligent dialogue technology, and in particular to a method and system for flexibly constructing multi-round dialogues in business scenarios. Background Art

[0002] Intelligent dialogue systems are widely used in the fields of intelligent customer service, robots, automobiles, navigation, etc. Under the background of the overall transformation of the automotive field to intelligent manufacturing, the automotive industry is also undergoing a major change, and various standardized processes are gradually replaced by machines. Similarly, for automobile sales or after-sales service personnel, they will also experience a change. The emergence of intelligent customer service builds a bridge for better communication between enterprises and users. Long-term data accumulation will give intelligent customer service infinite wisdom. For example, CN201911166714.9 discloses a method and system for building a multi-round dialogue system based on business scenarios, which provides a multi-round dialogue process designed according to different business scenarios, and configures corpus and keywords for each user node in the multi-round dialogue process, and configures titles and words in each machine node; the corpus and keywords of each user node in the multi-round dialogue process are respectively trained according to text classification technology and rule matching. Intention judgment model and rule matching model. The method configures node corpus and then distributes it through node judgment intent. However, the node in the method only supports intent judgment, and cannot support some complex services or build more complex business logic. It is essentially a template configuration method, which configures the dialogue template in advance and returns the dialogue template information by matching the input corpus information. It is limited to some simple scenarios and does not support expansion. For example, CN 202110609449.8 discloses a human-computer multi-round dialogue method and device, which includes: obtaining a process node; obtaining an intent library, each intent library includes intent information and basic corpus; obtaining a sentence to be analyzed; comparing the similarity of the sentence to be analyzed with each basic corpus, thereby obtaining the similarity value of the sentence to be analyzed with each basic corpus; and obtaining the corresponding response action according to the identified intent information. Since the intent library can be associated with multiple process nodes, there is no need to configure multiple basic corpora for each process node. However, when the method configures the intent and calculates the similarity with the sentence to be analyzed, it is easy to cause errors such as multiple intent conflicts.

[0003] Therefore, how to flexibly configure complex dialogue scenarios to build complex processes is a problem that intelligent customer service needs to solve. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the purpose of the present invention is to provide a method and system for flexibly configuring and building multiple rounds of dialogues based on business scenarios, so as to solve the problem that the application scenarios of the prior art are relatively fixed / single and cannot achieve flexible switching of scenarios to meet the needs of complex scenario links.

[0005] To achieve the above object, the present invention adopts the following technical solution:

[0006] A method for flexibly constructing a multi-round dialogue in a business scenario, characterized in that a data flow is constructed by constructing a start node, a logic node, an end node and a connection line, so as to realize flexible construction of a multi-round dialogue in a business scenario;

[0007] Among them, the start node serves as the beginning of the entire process and supports the configuration of common mandatory end corpus for the entire conversation; the connection line between the start node and the logic judgment node constructs the data flow graph into a complex conversation scenario; the logic judgment node provides external services to provide functions to the outside world and supports connecting multiple complex conversation scenarios in series; the end node is used to summarize information and return the information the user wants to the user after processing.

[0008] Furthermore, the method for flexibly constructing a multi-round dialogue in a business scenario includes the following steps:

[0009] Step 1: Strip the business data flow diagram according to complex business scenarios and create intent modules in advance;

[0010] Step 2: Construct the data flow graph into a complex dialogue scenario through the start node, logic judgment node, end node and connection line;

[0011] Step 3: Provide external services to provide external functions and support the connection of multiple complex dialogue scenarios.

[0012] Furthermore, the rules within the connection line between the start node and other logical nodes are used as logical judgments; in various dialogue scenarios, relevant intentions need to be established in advance according to the business scenarios. The entire dialogue process can be triggered in the connection line by selecting the set intentions in advance, and it also supports setting corpus triggers in the connection line.

[0013] The related intent supports directly outputting user input through related corpus through the NGT intent search method; this method efficiently and quickly matches the most relevant intent of the user input through approximate nearest neighbor vector search in high-dimensional space.

[0014] By configuring the trigger corpus, when the similarity between the user input corpus and the corpus configured in the scenario reaches the set threshold, the subsequent subgraph is triggered; in the logical judgment node, by reproducing the nodes of the complex process, multiple rounds of rhetorical questioning corpus are edited in the logical node.

[0015] Furthermore, there is at least one logic judgment node, and the logic judgment node is edited according to the business logic to obtain the information provided by the user in the complex question-and-answer scenario. After the user triggers the intention, the logic node will obtain some information that the user may enter in the scenario; in the logic judgment node, the rhetoric returned to the user can be directly configured, which can be set by writing a custom script; or multiple inquiries can be made, and the valid information entered by the user can be recorded in the context, and then the node flow can be performed according to the information obtained.

[0016] Furthermore, semantic analysis is performed based on the content returned by the user in the logic judgment node, and the analysis work is supported by building semantic rules, requesting external resources and self-developed scripts. The semantic rules are constructed by analyzing a large number of conversation logs, and various data analysis methods are used to extract information with obvious distinguishing features from the logs, and semantic rules are constructed based on the information. The request for external resources is to directly parse the content input by the user by calling external services to obtain the necessary information. The self-developed script is to customize the editing script in the logic judgment node to support certain complex scenarios, and implement various complex functions through the script, including but not limited to building complex scheduled tasks.

[0017] In various dialogue scenarios, it is necessary to establish relevant intents in advance according to the business scenarios. In the connection line, you can select the intent set in step 1 in advance to trigger the entire dialogue process by selecting the intent trigger. At the same time, you can support setting the corpus trigger in the connection line. By configuring the trigger corpus, when the similarity between the user input corpus and the corpus configured in the scenario reaches the set threshold, the subsequent process of the dialogue flow will be entered. In the logical judgment node, by reproducing the nodes in the complex process and editing the corpus of multiple rounds of rhetorical questions in the logical judgment node, you can obtain the completion of a certain scenario from the user side, but the content returned by the user side may not directly meet the requirements, so it is necessary to perform semantic analysis on the content returned by the user in the logical judgment node, and support the analysis work by building semantic rules, requesting external resources and self-developed scripts. Semantic rules are constructed by analyzing a large number of conversation logs and using various data analysis methods to extract information with obvious distinguishing features from the logs, and then constructing semantic rules based on the information. At the same time, external resources can be directly requested and external services can be called to directly parse the user input content and obtain the necessary information. In supporting certain complex scenarios, custom editing scripts can be used in logical nodes to implement various complex functions through scripts, including but not limited to building complex scheduled tasks.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] 1. The present invention is based on a flexible configuration construction method for multiple rounds of dialogues in business scenarios. It extracts information through nodes and supports custom coding operations to complete scenario design. Theoretically, it supports all scenarios. It solves the problem that the application scenarios of the prior art are relatively fixed and flexible switching of scenarios cannot be achieved to meet the needs of complex scenario links.

[0020] 2. The present invention uses connecting lines to make subsequent logical judgments, which is clearer and can quickly and intuitively establish dialogue flows for various complex scenarios.

[0021] 3. In addition to supporting the above methods, the intent matching method proposed in the present invention also supports customizing other judgment methods for verification while judging the intent, such as regular expressions. The intent judgment logic can only be determined to be triggered when multiple conditions are met at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A data flow chart of an embodiment of a method for flexibly configuring and building multiple rounds of dialogue based on business scenarios of the present invention;

[0023] Figure 2 This is a simplified logic diagram of the flexible configuration construction method and system based on business scenario multi-round dialogue of the present invention;

[0024] Figure 3 This is a logical flow node slot extraction diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described below in conjunction with the accompanying drawings and specific embodiments, but the embodiments of the present invention are not limited thereto.

[0026] A method for flexibly constructing multi-round dialogues in business scenarios includes: constructing a data flow by constructing a start node, a logic node, an end node and a connecting line, so as to realize flexible construction of multi-round dialogues in business scenarios.

[0027] See also Figure 1 , is a data flow diagram abstracted from a specific instance. The flow chart includes a start node, an end node, and two logic judgment nodes. Information transmission and condition judgment between nodes are performed through connecting lines.

[0028] Among them, the start node serves as the beginning of the entire process and supports the configuration of common mandatory end corpus for the entire conversation; the connection line between the start node and the logic judgment node constructs the data flow graph into a complex conversation scenario; the logic judgment node provides external services to provide functions to the outside world and supports connecting multiple complex conversation scenarios in series; the end node is used to summarize information and return the information the user wants to the user after processing.

[0029] See also Figure 2 , is a simplified logic diagram of the present invention, which abstractly represents the necessary steps to construct a dialogue scene. In order to achieve flexible scene switching, meet complex scene links, and solve the problem of single scene in the prior art, the present invention adopts the method of nodes and connecting lines to construct various complex dialogue processes, which specifically includes the following steps:

[0030] Step 1: First, strip the business data flow diagram according to a complex business scenario and create an intent module in advance;

[0031] Step 2: Use nodes and connecting lines to construct a data flow graph into a complex dialogue scenario;

[0032] Step 3: Provide external services to provide external functions and support the connection of multiple complex dialogue scenarios.

[0033] The purpose of creating the intent module in advance in step 1 is to provide a method for NGT intent search, which supports directly outputting the relevant intent of the user input through the relevant corpus. This method searches for the approximate nearest neighbor vector in the high-dimensional space to efficiently and quickly match the most relevant intent of the user input.

[0034] The specific implementation method of step 2 is: construct a data flow diagram by constructing a start node, a logic judgment node, an end node, and a connection line. First, create a start node, which serves as the beginning of the entire process and supports the configuration of common mandatory end corpus for the entire conversation. The rules in the connection line between the start node and other logic nodes are used as logic judgments. The connection line connecting the start node generally sets a trigger condition for the conversation process.

[0035] In various dialogue scenarios, it is necessary to establish relevant intents in advance according to the business scenarios. In the connection line, you can select the intent set in step 1 in advance to trigger the entire dialogue process by selecting the intent trigger method. At the same time, you can support setting the corpus trigger in the connection line. By configuring the trigger corpus, when the similarity between the user input corpus and the corpus configured in the scenario reaches the set threshold, it will enter the subsequent flowchart. In the logic judgment node, by reproducing the nodes in the complex process and editing the corpus of multiple rounds of rhetorical questions in the logic judgment node, you can obtain a scene from the user side. However, the content returned by the user side may not directly meet the requirements, so it is necessary to perform semantic analysis on the content returned by the user in the logic judgment node, and support the analysis work by building semantic rules, requesting external resources and self-developed scripts. Building semantic rules analyzes a large number of dialogue logs and uses various data analysis methods to extract information with obvious distinguishing features in the logs, and builds semantic rules through information; at the same time, you can also directly request external resources and call external services to directly parse the content input by the user to obtain the necessary information; in the case of supporting certain complex scenarios, you can customize the editing script in the logic node to implement various complex functions through the script, including but not limited to building complex scheduled tasks.

[0036] The specific implementation of the present invention is further described below with respect to a complete dialogue scenario.

[0037] 1. Prepare trigger corpora for responses based on business scenarios. For example, in a car buying scenario, prepare corpora such as "I want to buy a car" and "How much is XX car?" that are directly related to the business scenario. You can create a new car buying intent based on the corpus and fully understand that the corpus input by the user belongs to the car buying scenario by training an intent classification model or simply building a text semantic matching model.

[0038] 2. Construct a start node. The start node can be used as a brief description of the entire conversation scene. You can also set a forced interruption corpus at the start node to force the end of the entire complex conversation process. The specific implementation method is to determine whether the query is a forced end corpus, reset the conversation state directly, change the type of the conversation behavior to the end state, and return the scene end script preset in the start node.

[0039] 3. Create scenario trigger conditions (edit flow lines). Scenario triggers can be implemented in a variety of ways, including but not limited to intent recognition, corpus triggering, keyword verification, etc. Intent recognition is to call a pre-trained intent classification model to determine which intent best matches the user's input question, and then flow to the subsequent nodes; corpus triggering refers to editing the relevant trigger corpus in the flow line, and then using a pre-trained semantic model to determine whether the user input and the trigger text are similar; keyword verification mainly provides a verification method through regular rules to quickly determine whether the user input meets the conditions. The flow line is mainly used to judge the flow conditions. The subsequent flow situation and reply words can be set in advance in the flow line, and then flexibly respond according to the user input.

[0040] 4. Edit the logic judgment node according to the business logic. The logic judgment node is mainly used to obtain the information provided by the user in complex question-and-answer scenarios. For example, in a car buying scenario, after the user triggers the intention to buy a car, the logic node will obtain some information that the user may enter in the scenario, including but not limited to the intended vehicle, budget, etc. In the logic judgment node, you can directly configure the rhetoric returned to the user, or you can set it by writing a custom script, such as calculating and filtering the price based on the information entered by the user, or you can ask multiple times, and record the valid information entered by the user in the context, and then transfer the nodes according to the information obtained.

[0041] 5. End node. The purpose of the end node is to summarize information and return the information the user wants to the user after processing. For example, in the scenario where the user buys a car, what may be returned to the user is a comparison of various pre-selected models, but now almost all platforms do not directly support it. However, through the method of the present invention, the data of each model is obtained in the end node and the page is customized in the desired format. The end node supports multiple encoding methods, and the main implementation method is code embedding.

[0042] See also Figure 3 , which is a partial enlarged view of the logic node in the dialogue flow chart. Figure 3 It represents the extraction of key information from the conversation data. When the user interacts with the system, the system will parse the conversation information and extract the coordinates or longitude and latitude information from the conversation information through custom coding to complete the execution of the subsequent conversation flow.

[0043] It can be seen that the method of the present invention can solve the problem of difficulty in constructing multiple rounds of question and answer in complex scenarios.

[0044] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit the technical solution. Those skilled in the art should understand that those modifications or equivalent substitutions of the technical solution of the present invention that do not depart from the purpose and scope of the technical solution should be included in the scope of the claims of the present invention.

Claims

1. A method for flexibly constructing multi-round dialogues in business scenarios, characterized in that: Including constructing start nodes, logic judgment nodes, end nodes and connecting lines to build data flow; The start node is the beginning of the entire process and supports the configuration of the common mandatory end corpus of the entire conversation; the connection line between the start node and the logic judgment node constructs the data flow graph into a complex conversation scene; the logic judgment node provides external services and supports connecting multiple complex conversation scenes; the end node is used to summarize information and return the information the user wants to the user after processing; The rules in the connection line between the start node and other logical nodes are used as logical judgments; in various dialogue scenarios, relevant intentions need to be established in advance according to the business scenarios. The entire dialogue process can be triggered by selecting the set intentions in advance in the connection line, and the setting of corpus triggers in the connection line is also supported; After the user triggers the intent, the logic judgment node will obtain some information that the user may enter in the scene; in the logic judgment node, you can directly configure the rhetorical questions returned to the user, which can be set by writing a custom script; or you can ask multiple times and record the valid information entered by the user in the context, and then transfer nodes based on the information obtained.

2. The method for flexibly constructing multi-round dialogues in business scenarios according to claim 1, characterized in that: The following steps are involved: Step 1: Strip the business data flow diagram according to complex business scenarios and create intent modules in advance; Step 2: Construct the data flow graph into a complex dialogue scenario through the start node, logic judgment node, end node and connection line; Step 3: Provide external services to provide external functions and support the connection of multiple complex dialogue scenarios.

3. The method for flexibly constructing multi-round dialogues in business scenarios according to claim 1 or 2, characterized in that: There is at least one logic judgment node, which is edited according to the business logic to obtain information provided by the user in a complex question-and-answer scenario.

4. The method for flexibly constructing multi-round dialogues in business scenarios according to claim 2, characterized in that: The related intent module supports direct output of the related intent of the user input by inputting related corpus through the high-dimensional vector retrieval algorithm - NGT intent search; this method efficiently and quickly matches the most relevant intent of the user input through the approximate nearest neighbor vector search in the high-dimensional space.

5. The method for flexibly constructing multi-round dialogues in business scenarios according to claim 1, characterized in that: By configuring the trigger corpus, when the similarity between the user input corpus and the corpus configured in the scenario reaches the set threshold, the subsequent subgraph is triggered; in the logical judgment node, by reproducing the nodes of the complex process, multiple rounds of rhetorical questioning corpus are edited in the logical node.

6. The method for flexibly constructing multi-round dialogues in business scenarios according to claim 1, characterized in that: Semantic analysis is performed based on the content returned by the user in the logical judgment node, and the analysis work is supported by building semantic rules, requesting external resources and self-developed scripts.

7. The method for flexibly constructing multi-round dialogues in business scenarios according to claim 6, characterized in that: The semantic rules are constructed by analyzing a large number of conversation logs and extracting information with obvious distinguishing features from the logs through various data analysis methods, and the semantic rules are constructed based on the information; the external resource is requested by calling external services to directly parse the content input by the user and obtain the necessary information.

8. The method for flexibly constructing multi-round dialogues in business scenarios according to claim 6, characterized in that: The self-developed script is to customize the editing script in the logic judgment node to support certain complex scenarios, and implement various complex functions through the script, including but not limited to building complex scheduled tasks.

9. A system for flexibly constructing multi-round dialogues in business scenarios, characterized in that: Adopt any method of claim 1-8.

Images