Customer service method and system based on two layers of knowledge maps

Abstract

The invention discloses a customer service method and system based on a two-layer knowledge graph, relates to the technical field of knowledge graphs, and solves the technical problem that an existing customer service system can only reply some simple consultation problems by utilizing the knowledge graph and cannot process complex consultation. The method comprises the following steps: constructing atomic force groups, wherein each atomic force group consists of a plurality of basic nodes; after the atomic power group is created, extracting to-be-combined atomic power group nodes for combination, and combining to obtain a customer service map; obtaining a query intention of the user according to the query voice of the user; outputting a service scheme corresponding to the query intention based on the customer service map; according to the invention, a corresponding service scheme can be provided for the query of the user based on the constructed customer service map, the service efficiency and flexibility are improved, and better and more efficient service experience is brought to the user.

Description

Technical Field

[0001] The present invention belongs to the field of knowledge graphs, and specifically relates to a customer service method and system based on a two-layer knowledge graph. Background Art

[0002] In customer service, knowledge graphs were initially used to handle some simple consulting questions, relying on parsing user questions and retrieving relevant information from pre-built knowledge graphs to provide answers. This approach greatly improves response speed and accuracy, especially in handling high-frequency and standardized questions.

[0003] However, existing technologies are unable to cope with more complex customer issues. Complex problems usually involve multiple requirements or conditions, such as the need to query information across multiple systems, combine historical interaction records for personalized suggestions, or resolve non-standard service requests. Currently, knowledge graphs can only be used to answer some simple consulting questions, and their application effect is not ideal for these more complex scenarios. The main reasons include but are not limited to: Limited coverage of knowledge graphs: Existing knowledge graphs may not cover all necessary entities and relationships, especially those details that are specific to certain business processes or customer needs. Lack of deep reasoning capabilities: Although knowledge graphs can organize and express information well, existing technical solutions often cannot provide sufficient support when faced with problems that require in-depth analysis and logical reasoning. Therefore, the present invention provides a customer service method and system based on a two-layer knowledge graph. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a customer service method and system based on a two-layer knowledge graph, which is used to solve the technical problem that the existing customer service system can only respond to some simple consulting questions using the knowledge graph but cannot handle complex consulting questions.

[0005] To achieve the above object, the first aspect of the present invention provides a customer service method based on a two-layer knowledge graph, comprising the following steps:

[0006] Step 1: construct atomic capability groups based on historical data; wherein the atomic capability groups are several service themes; each atomic capability group is composed of several basic nodes; the historical data is the customer service consulting service content corresponding to the several service themes;

[0007] Step 2: Extract the atomic capability groups to be combined and combine them to obtain a customer service graph;

[0008] Step 3: Obtain the user's query intention based on the user's query voice; output the service plan corresponding to the query intention based on the customer service graph.

[0009] Preferably, the construction of the atomic capability group includes:

[0010] Extract several service contents of customer service from historical data, and divide the extracted service contents into several service themes according to the content themes; extract query information related to the service themes from historical data; determine several basic nodes in each service theme based on the query information; combine several basic nodes to obtain an atomic capability group; wherein the basic nodes include judgment nodes, call interface nodes, service script question nodes and service script reply nodes.

[0011] Preferably, the atomic capability group has one request receiving point, and one request receiving point corresponds to several response results.

[0012] Preferably, the method for obtaining the atomic capability group to be combined includes:

[0013] Based on the service content of each atomic capability group, extract the key words in the service content of each atomic capability group, convert the key words into word vectors, and calculate the cosine similarity between the word vectors of each atomic capability group; judge whether the cosine similarity between the word vectors is greater than the preset similarity threshold; if yes, the atomic capability groups are similar and marked as atomic capability groups to be combined; if no, the atomic capability groups are not similar and no marking is done.

[0014] Preferably, combining the nodes of the atomic capability group to be combined includes:

[0015] The atomic capability groups to be combined are grouped into a layer of graph; wherein the layer of graph does not display the node details in the atomic capability group;

[0016] Each atomic capability group in the first-layer graph is expanded to obtain a second-layer graph; the second-layer graphs of several atomic capability groups constitute a customer service graph.

[0017] Preferably, the step of expanding each atomic capability group in a layer of the atlas includes:

[0018] A directional edge is drawn out from the atomic capability group to be expanded in one layer of the graph, and points to the entrance of the next atomic capability group, and the exit of the next atomic capability group is connected to the atomic capability group in the next layer of the graph.

[0019] Preferably, obtaining the user's query intention according to the user's query voice includes:

[0020] The user's query voice and the corresponding query intent label are input into the intent recognition model, the intent recognition model outputs the intent label, the intent label is matched with the preset intent, and the query intent corresponding to the user's query voice is obtained; wherein, the intent recognition model is built based on the NLP intent recognition model.

[0021] Preferably, the construction process of the intention recognition model is as follows, including:

[0022] Extract query voices and query intentions of several users from historical data;

[0023] Integrate several query voices and corresponding query intentions into several groups of training data and test data; use the training data to train the NLP intent recognition model; use the test data to test the trained NLP intent recognition model; adjust the NLP intent recognition model according to the test results; and finally obtain an intent recognition model whose input is query voice and output is query intention.

[0024] Preferably, based on the customer service graph, the process of outputting the corresponding service plan is as follows, including:

[0025] Step S1: Determine whether this execution is a breakpoint execution; if yes, execute directly from the basic node where the last execution was interrupted; if no, execute from the starting basic node;

[0026] Step S2: When the execution reaches the service script question node, the service script will be fed back to the front-end user, and the execution will continue after the user replies; when the execution reaches the call interface node, the background interface will be directly called to query the cache information and return the interface call result; key information is obtained from the information returned by the interface and the user's reply, and jumps to the next basic node until the service plan corresponding to the user's intention is obtained and the execution is completed.

[0027] Preferably, the second aspect of the present invention provides a customer service system based on a two-layer knowledge graph, including a customer service graph construction module and a query module;

[0028] Customer service graph construction module: used to construct atomic capability groups, extract the nodes of the atomic capability groups to be combined, and combine them to obtain the customer service graph;

[0029] Query module: used to obtain the user's query intention based on the user's query voice; based on the customer service map, output the service plan corresponding to the query intention.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] The present invention splits the complex customer service into several atomic capabilities, so that each atomic capability group has independent and clear functions, improves the scalability and flexibility of the system, and because each atomic capability is independent, it can be optimized or upgraded for a specific node without affecting the rest of the entire system; according to the service theme, select appropriate nodes from the constructed atomic capability group for combination to form a diversified customer service map, which can adapt to the service needs in different scenarios. Through reasonable node combination, effective coordination between different atomic capability groups can be achieved to improve service efficiency and quality; through speech recognition and natural language processing technology, the customer service map can accurately understand the user's query intention, and quickly output the corresponding service plan according to the customer service map, and the intelligent service improves the user's satisfaction and experience. In addition, the two-layer map structure can efficiently reuse the atomic capability group and reduce the difficulty of drawing. For example, a complete second-layer map has one hundred nodes, and a layer of map has only about ten nodes. Using a two-layer map structure, only a map of about ten nodes is needed to improve drawing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0033] Figure 1 It is a schematic diagram of the process of the present invention;

[0034] Figure 2 A schematic flow chart of a method for determining the atomic capability group to be combined according to the present invention;

[0035] Figure 3 This is a flowchart of the process of expanding the first-layer graph to the second-layer graph of the present invention;

[0036] Figure 4 This is a partial two-layer diagram of the present invention. DETAILED DESCRIPTION

[0037] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] See also Figure 1 The first embodiment of the present invention provides a customer service method based on a two-layer knowledge graph, comprising the following steps:

[0039] Step 1: Build atomic capability groups;

[0040] Specifically, several service contents of customer service are extracted from historical data, and the extracted service contents are divided into several service themes according to the content themes; query information related to the service themes is extracted from historical data; several basic nodes in each service theme are determined based on the query information; several basic nodes are combined to obtain an atomic capability group; wherein the service content is the content of several user queries stored in the database.

[0041] Among them, the service content is the content of several user queries stored in the database; among them, each atomic capability group is composed of several basic nodes; the atomic capability group is several service topics; the basic nodes include judgment nodes, call interface nodes, service script question nodes and service script reply nodes; each atomic capability group represents a complete function.

[0042] It should be noted that one atomic capability group has one request receiving point, and one request receiving point corresponds to several response results.

[0043] Each atomic capability group can only have one entry, that is, an edge entering the node within the atomic capability group. If the program may end at this node, then the node needs to have an edge pointing to the end node.

[0044] For example, suppose that the problem that users of a broadband service center often inquire about is that the broadband has no signal. The service theme is determined as follows: broadband fault handling; judgment node: determine whether the user has subscribed to the broadband service, etc.; call API capability node: query and display the details of the products and services that the customer has subscribed to, such as calling the interface to query the broadband information under the package, and obtain the broadband account and address; speech question node: confirm whether the faulty broadband account can be provided, such as "In order to better help you, please provide the faulty broadband account." and "Please tell us the mobile phone number bound to the faulty broadband for verification." etc.; speech reply node: provide corresponding replies based on the information provided by the user and the execution results of the above nodes. For example, if the user's broadband account cannot be found, you can reply: "Sorry, we cannot find information that matches the provided broadband account. Please confirm again or provide more details." Several basic nodes of this broadband processing constitute the atomic capability group about "solving customer broadband fault problems."

[0045] Step 2: After the atomic capability group is created, extract the nodes of the atomic capability group to be combined and combine them to obtain the customer service graph;

[0046] See also Figure 2Specifically, based on the service content of each atomic capability group, extract the key words in the service content of each atomic capability group, convert the key words into word vectors, and calculate the cosine similarity between the word vectors of each atomic capability group; determine whether the cosine similarity between the word vectors is greater than a preset similarity threshold; if yes, the atomic capability groups are similar and marked as atomic capability groups to be combined; if no, the atomic capability groups are not similar and are not marked.

[0047] For example, suppose there are two atomic capability groups. Atomic capability group 1: new user broadband application; service content: determine whether the provided mobile phone number has information, confirm whether the installation address can be provided, and call the interface to query available broadband packages; keywords: mobile phone number, installation address, available broadband packages;

[0048] Atomic capability group 2: broadband upgrade for existing users; service content: determine whether the customer has broadband, confirm the current broadband package, and call the interface to query the upgradeable broadband package; keywords: mobile phone number, installation address, current broadband package, upgrade broadband package;

[0049] Assume that a simple bag-of-words model is used to represent these keywords. The vocabulary includes all the mentioned keywords: "mobile phone number", "installation address", "available broadband packages", "broadband", "current broadband packages", "upgrade broadband packages". Each atomic capability group is represented as a vector; where each dimension represents the number of times a word appears in the vocabulary. The word vector of atomic capability group 1: V1 = [1,1,1,0,0,0]; the word vector of atomic capability group 2: V2 = [1,1,0,0,1,1]; calculate the cosine similarity between atomic capability group 1 and atomic capability group 2 = (V1×V2) / (‖V1‖×‖V2‖)≈0.577. Assuming the preset similarity threshold is 0.4, the two atomic capability groups are considered to be combined atomic capability groups.

[0050] The atomic capability groups to be combined are grouped into a layer of graph; wherein the layer of graph does not display the node details in the atomic capability group;

[0051] Each atomic capability group in the first-layer graph is expanded. Specifically, an edge is drawn from the first-layer graph of the current atomic capability group to point to the entrance of the atomic capability group. The exit of the atomic capability group is connected to the atomic capability group in the next first-layer graph to obtain a second-layer graph. The second-layer graphs of several atomic capability groups constitute a customer service graph.

[0052] See also Figure 3 , the original first-layer graph node remains unchanged, and an edge is drawn from the first-layer graph node to point to the entrance of the atomic capability group, and the exit of the atomic capability group is connected to the next original first-layer graph node. The Group_next edge represents the edge of the first-layer graph, and the next edge represents the edge of the second-layer graph. This structure combines the first-layer graph with the second-layer graph.

[0053] See also Figure 4 , part of the structure of the two-layer graph is given. The judgment nodes in the graph include "judging whether the incoming call number has information", "judging whether the customer has broadband", "judging whether the provided account has information", "judging whether the customer has broadband", "judging whether the provided mobile phone number has information", "judging whether there are two broadbands under the package", "judging whether there is only one broadband under the package"; the calling interface nodes include "judging the incoming call number information on the customer's homepage", "same as the customer's homepage product information area", "calling the interface to query the broadband information under the package and obtain the broadband account and address"; the questioning nodes include "confirming whether the faulty broadband account can be provided", "confirming whether the mobile phone number bound to the faulty broadband can be provided", "confirming whether the broadband is handled under the incoming call number certificate".

[0054] Step 3: Obtain the user's query intention based on the user's query voice; output the service plan corresponding to the query intention based on the customer service graph.

[0055] Specifically, the user's query voice and the corresponding query intention label are input into the intent recognition model, the intent recognition model outputs the intention label, the intention label is matched with the preset intent, and the query intent corresponding to the user's query voice is obtained; wherein, the intent recognition model is constructed based on the NLP intent recognition model.

[0056] The process of building the intent recognition model is as follows:

[0057] Extract query voices and query intentions of several users from historical data;

[0058] Integrate several query voices and corresponding query intentions into several groups of training data and test data; use the training data to train the NLP intent recognition model; use the test data to test the trained NLP intent recognition model; adjust the NLP intent recognition model according to the test results; and finally obtain an intent recognition model whose input is query voice and output is query intention.

[0059] Based on the customer service graph, the process of outputting the corresponding service plan is as follows:

[0060] Step S1: Determine whether this execution is a breakpoint execution; if yes, execute directly from the basic node where the last execution was interrupted; if no, execute from the starting basic node;

[0061] Step S2: If the execution reaches the service script question node, the service script will be fed back to the front-end user, and the execution will continue after the user replies; if the execution reaches the call interface node, the background interface will be directly called to query the cache information and return the interface call result; key information will be obtained from the information returned by the interface and the user's reply, and jump to the next basic node until the service plan corresponding to the user's intention is obtained and the execution is completed.

[0062] For example, for the intent "broadband / whole-house WIFI cannot access the Internet", the corresponding service plan is obtained, and the service plan is a complete two-layer map.

[0063] A second aspect of the present invention provides a customer service system based on a two-layer knowledge graph, including a customer service graph construction module and a query module;

[0064] Customer service graph construction module: used to construct atomic capability groups, extract the nodes of the atomic capability groups to be combined, and combine them to obtain the customer service graph;

[0065] Query module: used to obtain the user's query intention based on the user's query voice; based on the customer service map, output the service plan corresponding to the query intention.

[0066] Part of the data in the above formula is calculated by removing the dimension and taking its numerical value. The formula is a formula closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0067] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A customer service method based on a two-layer knowledge graph, characterized in that: include: Construct atomic capability groups based on historical data; wherein the atomic capability groups are several service themes; each atomic capability group is composed of several basic nodes; the historical data is the customer service consulting service content corresponding to the several service themes; Extract the atomic capability groups to be combined and combine them to obtain a customer service graph; Obtain the user's query intention based on the user's query voice; Based on the customer service graph, output the service plan corresponding to the query intent.

2. The customer service method based on two-layer knowledge graph according to claim 1 is characterized in that: The atomic capability group is constructed, including: Extract several service contents of customer service from historical data, and divide the extracted service contents into several service themes according to the content themes; extract query information related to the service themes from historical data; determine several basic nodes in each service theme based on the query information; combine several basic nodes to obtain an atomic capability group; wherein the basic nodes include judgment nodes, call interface nodes, service script question nodes and service script reply nodes.

3. The customer service method based on two-layer knowledge graph according to claim 2 is characterized in that: The atomic capability group has one request receiving point, and one request receiving point corresponds to several response results.

4. The customer service method based on two-layer knowledge graph according to claim 2 is characterized in that: The method for obtaining the atomic capability group to be combined includes: Based on the service content of each atomic capability group, extract the key words in the service content of each atomic capability group, convert the key words into word vectors, and calculate the cosine similarity between the word vectors of each atomic capability group; determine whether the cosine similarity between the word vectors is greater than the preset similarity threshold; if yes, mark the corresponding atomic capability group as the atomic capability group to be combined; if not, do not mark the corresponding atomic capability group.

5. The customer service method based on two-layer knowledge graph according to claim 4 is characterized in that: Combine the atomic capability groups to be combined, including: The atomic capability groups to be combined are grouped into a layer of graph; wherein the layer of graph does not display the node details in the atomic capability group; Each atomic capability group in the first-layer graph is expanded to obtain a second-layer graph; the second-layer graphs of several atomic capability groups constitute a customer service graph.

6. The customer service method based on two-layer knowledge graph according to claim 5 is characterized in that: The expansion of each atomic capability group in a layer of the atlas includes: A directional edge is drawn out from the atomic capability group to be expanded in one layer of the graph, and points to the entrance of the next atomic capability group, and the exit of the next atomic capability group is connected to the atomic capability group in the next layer of the graph.

7. The customer service method based on two-layer knowledge graph according to claim 6 is characterized in that: The obtaining of the user's query intention according to the user's query voice includes: The user's query voice and the corresponding query intent label are input into the intent recognition model, the intent recognition model outputs the intent label, the intent label is matched with the preset intent, and the query intent corresponding to the user's query voice is obtained; wherein, the intent recognition model is built based on the NLP intent recognition model.

8. The customer service method based on two-layer knowledge graph according to claim 7 is characterized in that: The construction process of the intent recognition model is as follows, including: Extract query voices and query intentions of several users from historical data; Integrate several query voices and corresponding query intentions into several groups of training data and test data; use the training data to train the NLP intent recognition model; use the test data to test the trained NLP intent recognition model; adjust the NLP intent recognition model according to the test results; and finally obtain an intent recognition model whose input is query voice and output is query intention.

9. The customer service method based on two-layer knowledge graph according to claim 8 is characterized in that: Based on the customer service graph, the process of outputting the corresponding service plan is as follows, including: Step S1: Determine whether this execution is a breakpoint execution; if yes, execute directly from the basic node where the last execution was interrupted; if no, execute from the starting basic node; Step S2: When the execution reaches the service script question node, the service script will be fed back to the front-end user, and the execution will continue after the user replies; when the execution reaches the call interface node, the background interface will be directly called to query the cache information and return the interface call result; key information is obtained from the information returned by the interface and the user's reply, and jumps to the next basic node until the service plan corresponding to the user's intention is obtained and the execution is completed.

10. A customer service system based on a two-layer knowledge graph, operating based on a customer service method based on a two-layer knowledge graph as described in any one of claims 1 to 9, characterized in that: Includes customer service graph building module and query module; Customer service graph construction module: used to construct atomic capability groups, extract the nodes of the atomic capability groups to be combined, and combine them to obtain the customer service graph; Query module: used to obtain the user's query intention based on the user's query voice; Based on the customer service graph, output the service plan corresponding to the query intent.

Images