An intelligent question and answer implementation method and device, electronic equipment and storage medium

Abstract

Embodiments of the present application disclose a kind of intelligent question and answer implementation method, device, electronic equipment and storage medium.The method comprises: obtaining the question information of user input, and the question information is converted into target information;Target information is input to the non-uniform graph neural network model trained in advance, and the output information of non-uniform graph neural network model is obtained;The output information is converted into the answer corresponding to question information, and the answer is shown to user.The method of the present application embodiment can utilize the non-uniform graph neural network model established in advance, analyze the question information of user, and accurately and quickly provide the answer information corresponding to question for user.The speed of processing question is fast, and the accuracy of output answer is high, further improve user experience.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the field of intelligent question answering, and in particular to an intelligent question answering implementation method and device, electronic equipment and a storage medium. BACKGROUND

[0002] Intelligent question answering is to sort accumulated disordered corpus information in order and scientifically, and to establish a classification model based on knowledge; these classification models can guide newly added corpus consultation and service information, save human resources, improve the automaticity of information processing, and reduce website operation costs. Nowadays, with the development of science and technology, intelligent question answering has been more and more widely applied in people's life.

[0003] Existing intelligent question answering technologies include intelligent question answering based on template matching algorithm and intelligent question answering based on traditional deep learning. The intelligent question answering based on template matching algorithm has a higher requirement for personnel constructing data sets, and the personnel need to have in-depth research on a certain professional field to construct high-quality templates, and has a higher requirement for the size of the templates, which is time-consuming and labor-intensive, and the performance of the constructed intelligent question answering system is also poor. The intelligent question answering based on traditional deep learning method does not have a good processing method for the non-uniform distribution problem of the data set, and the data in the real scene is often non-uniformly distributed, which leads to poor intelligent question answering effect based on traditional deep learning method and low accuracy of the generated answers. SUMMARY

[0004] The present application provides an intelligent question answering implementation method, device, electronic equipment and storage medium, which can obtain the answers to the questions by using the non-uniform graph neural network model, and has high speed and high accuracy in processing questions.

[0005] In a first aspect, the embodiments of the present application provide an intelligent question answering implementation method, which comprises:

[0006] obtaining question information input by a user and converting the question information into target information;

[0007] inputting the target information into a pre-trained non-uniform graph neural network model to obtain output information of the non-uniform graph neural network model;

[0008] converting the output information into an answer corresponding to the question information, and displaying the answer to the user.

[0009] In a second aspect, the embodiments of the present application further provide an intelligent question answering implementation device, which comprises:

[0010] an obtaining module configured to obtain question information input by a user and convert the question information into target information;

[0011] an input module, configured to input the target information into a pre-trained non-uniform graph neural network model to obtain output information of the non-uniform graph neural network model;

[0012] an output module, configured to convert the output information into an answer corresponding to the question information and display the answer to the user.

[0013] In a third aspect, an electronic device is provided, and the electronic device comprises:

[0014] one or more processors;

[0015] a memory configured to store one or more programs;

[0016] When the one or more programs are executed by the one or more processors, the one or more processors implement the intelligent question answering implementation method provided in any embodiment of the present application.

[0017] In a fourth aspect, a computer readable storage medium is provided, and the computer readable storage medium stores a computer program, which, when executed by a processor, implements the intelligent question answering implementation method provided in any embodiment of the present application.

[0018] In the embodiments of the present application, question information input by a user is obtained, and the question information is converted into target information; the target information is input into a pre-trained non-uniform graph neural network model to obtain output information of the non-uniform graph neural network model; and the output information is converted into an answer corresponding to the question information and displayed to the user. That is, in the embodiments of the present application, the pre-established non-uniform graph neural network model is used to analyze question information of a user, and accurate and fast answer information corresponding to the question is output to the user, the speed of processing the question is fast, the accuracy of the output answer is high, and the user experience is further improved. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a flowchart of the intelligent question answering implementation method provided in the embodiments of the present application;

[0020] Figure 2 is a flowchart of the training method of the non-uniform graph neural network model provided in the embodiments of the present application;

[0021] Figure 3 is a schematic diagram of a node-edge structure provided in the embodiments of the present application;

[0022] Figure 4a is a node-edge structure schematic diagram of a question and answer sample library provided in the embodiments of the present application;

[0023] Figure 4bis a schematic diagram of a node-edge structure after edges with a degree less than 3 are deleted, provided by an embodiment of the present application;

[0024] Figure 5 is a structural schematic diagram of an intelligent question answering implementation device, provided by an embodiment of the present application;

[0025] Figure 6 is a structural schematic diagram of an electronic device, provided by an embodiment of the present application. DETAILED DESCRIPTION

[0026] The present application will be further described below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely intended for the purpose of interpretation of the present application and are not limiting of the present application. In addition, it should be noted that, for the purpose of description, only the parts related to the present application are shown in the accompanying drawings and not all the structures.

[0027] Figure 1 is a flowchart of an intelligent question answering implementation method, provided by an embodiment of the present application. The method of the present embodiment can obtain answers to questions by using a non-uniform graph neural network model, and has a high speed and high accuracy in processing questions. The method can be executed by an intelligent question answering implementation device of an embodiment of the present application, which can be integrated in an electronic device, which can be a server. The method can be implemented in a software and / or hardware manner. The intelligent question answering implementation method provided by the present embodiment specifically includes the following steps:

[0028] Step 101, obtaining question information input by a user and converting the question information into target information.

[0029] The question information is question information sent by the user to the server based on the user's own needs. The target information is information that can be input into a pre-trained non-uniform graph neural network model and corresponds to the question information input by the user, such as a feature vector of the question information.

[0030] In an optional implementation, the question information input by the user can be text information or voice information. When the question information is text information, the server can convert the text information into a feature vector (target information) by using a word embedding model after receiving the text information sent by the user. Commonly used word embedding models include vector space models, topic models, distributed models, deep pre-training models, and text representation models under specific tasks, etc. When the question information is voice information, the server can convert the received voice information into text information. Further, the text information is further processed to obtain the target information.

[0031] Step 102, inputting the target information into a pre-trained non-uniform graph neural network model to obtain output information of the non-uniform graph neural network model.

[0032] The non-uniform graph neural network model is a kind of graph neural network model. The graph neural network model refers to a network model using a neural network to learn graph structure data, extract and mine features and patterns in the graph structure data, and meet the needs of clustering, classification, prediction, segmentation, generation and other graph learning tasks. The graph neural network model can effectively extract key information of data by using the interaction between data. Compared with the traditional neural network model, the graph neural network model can more effectively process unstructured data such as text information. The non-uniform graph neural network model can be used to process unstructured data, which refers to data with non-uniform distribution problems. The non-uniform graph neural network model can further process unstructured data by using key elements in the data.

[0033] In an optional implementation, after obtaining the target information, the server can input the target information into the pre-trained non-uniform graph neural network model. The trained non-uniform graph neural network model can process the target information to obtain the output information corresponding to the target information. Therefore, before inputting the target information into the non-uniform graph neural network model, the initial graph neural network model can be trained. In the embodiment of the present scheme, the training of the initial graph neural network model to obtain the non-uniform graph neural network model includes the following steps A1-A2:

[0034] Step A1: If the non-uniform graph neural network model does not meet the pre-set convergence condition, a sample in the question and answer sample library is extracted as a current sample.

[0035] The question and answer sample library includes nodes, edges, and position information of the nodes in the pre-determined anchor structure. The convergence condition can be pre-set according to specific requirements and actual training process. In the present scheme, the convergence condition of the non-uniform graph neural network model is that the error between the sample data and the sample label is less than the pre-set error value. The question and answer sample library is composed of the Dilan ticket and the Dilan knowledge base. The Dilan ticket includes various question-answer form text information, and the Dilan knowledge base is an information base including big data and various public knowledge. The nodes are nouns in the Dilan ticket and the Dilan knowledge base, and the edges are the relationships between the nouns. The anchor structure is a structure pre-determined according to key edges and key points.

[0036] The question and answer sample library includes a large number of samples. In an optional implementation, when the non-uniform graph neural network model does not meet the pre-set convergence condition, the question and answer sample library is obtained and a sample in the question and answer sample library is extracted as a current sample.

[0037] Step A2: The current sample is used to train the non-uniform graph neural network model until the non-uniform graph neural network model meets the convergence condition.

[0038] Specifically, after the current sample is determined, the current sample is input into the non-uniform graph neural network, and output data corresponding to the current sample is obtained. Further, a loss function of the non-uniform graph neural network model is determined according to the output data and the sample label, and the network parameters of the non-uniform graph neural network model are adjusted according to the calculation result of the loss function until the non-uniform graph neural network model meets a convergence condition.

[0039] In the above steps, the initial graph neural network model can be trained by a large number of existing question and answer information libraries to obtain a trained non-uniform graph neural network model for processing problem information. The accuracy of the output answer information is improved, and the user experience is further improved.

[0040] Step 103, converting the output information into an answer corresponding to the problem information and displaying the answer to the user.

[0041] The output information is information output by the trained non-uniform graph neural network model after the target information is input into the trained non-uniform graph neural network model. Specifically, after the input information is input into the non-uniform graph neural network model, the non-uniform graph neural network model can output output information corresponding to the input information after calculation. After obtaining the output information, the output information can be converted into answer information corresponding to the problem information by using a natural language processing model.

[0042] For example, if the problem information input by the user is text information, the output information of the non-uniform graph neural network model can be converted into a text form answer after the output information is obtained, and the text form answer is sent to the user. When the problem information input by the user is voice information, the output information of the non-uniform graph neural network model is converted into a text answer after the output information is obtained. Further, the text form answer information is converted into a voice answer, and the voice answer is played for the user.

[0043] The technical scheme of the embodiment of the present application can analyze the problem information of the user by using the pre-established non-uniform graph neural network model, accurately and quickly output the answer information corresponding to the problem, process the problem quickly, and has high accuracy of the output answer, thereby improving the user experience.

[0044] Figure 2 is a flowchart of the training method of the non-uniform graph neural network model provided by the embodiment of the present application, as shown in the figure, the method mainly includes the following steps: Figure 2 ​

[0045] Step 201, determine the question and answer sample library of the initial graph neural network model and the sample label corresponding to each sample in the question and answer sample library.

[0046] Wherein, the question and answer sample library includes nodes, edges and location information of the nodes in the question and answer sample library. The information in the question and answer sample library comes from the Dijing work order and the Dijing knowledge base. The initial graph neural network model is a graph neural network model that has not started training. The sample label is the idealized output information corresponding to the sample information in the question and answer sample library. In actual application, the data in the question and answer sample library often has the problem of uneven distribution. In order to overcome the problem of uneven distribution of data and improve the training effect of the non-uniform graph neural network model, it is necessary to determine the anchor structure in the question and answer sample library. In the embodiment scheme, before the samples in the question and answer sample library are input into the initial graph neural network, the anchor structure mainly includes the following steps B1-B5:

[0047] Step B1: obtain the node degree of all nodes in the question and answer sample library.

[0048] Wherein, each noun in the question and answer sample library is a node. The edge is the relationship between each node. The node degree of a node represents the number of edges associated with other nodes and the node. The more edges connected between a node and other nodes, the greater the node degree of the node. Specifically, after obtaining the question and answer sample library, the node degrees of all nodes in the question and answer sample library are counted.

[0049] Step B2: add or subtract 1 from the node degrees of the two nodes connected by each edge to obtain the edge degree of each edge.

[0050] Wherein, the node degree of a node represents the number of edges associated with other nodes and the node. The edge degree of each edge is related to the node degrees of the two nodes connected by the edge. The greater the node degrees of the two nodes connected by each edge, the greater the edge degree of the edge.

[0051] Specifically, after obtaining the node degrees of all nodes, the edge degrees of each edge are calculated according to the node degrees of each node. For example, Figure 3 is a schematic diagram of the node-edge structure provided by the embodiment of the present application. As Figure 3 shown, the node degree of node 1 is 2, the node degree of node 2 is 3, the node degree of node 3 is 1, the node degree of node 4 is 1, and the node degree of node 5 is 1. Therefore, the edge degree between node 1 and node 2 is 4.

[0052] Step B3: take the initial edge parameter value as the current edge degree; delete the edges in the question and answer sample library with an edge degree less than the current edge degree to obtain an initial key edge set.

[0053] The initial key edge set is an edge set composed of the remaining edges in the question and answer sample library, and the (current edge degree minus 1) degree edge set is an edge set composed of the deleted edges in the question and answer sample library. The edge parameter value is a pre-set parameter value. For example, the initial edge parameter value is 2, and the current edge degree is 2. Assuming that the question and answer sample library includes edges with degrees of 1, 2, 3, 4, and 5, the edge with a degree of 1 is deleted from the question and answer sample library, and the remaining edge set is the initial key edge set.

[0054] For example, Figure 4a is a node-edge structure diagram of the question and answer sample library provided by an embodiment of the present application. Figure 4b is a node-edge structure diagram after the edges with degrees less than 3 are deleted. As shown in Figure 4a , the edge between node 1 and node 2 has a degree of 4, the edge between node 2 and node 4 has a degree of 2, the edge between node 2 and node 5 has a degree of 2, the edge between node 1 and node 3 has a degree of 2, and the edge between node 3 and node 6 has a degree of 2. As shown in Figure 4b , after the edges with degrees less than 3 are deleted from the question and answer sample library, the nodes in the node-edge structure are only node 1, node 2, node 3, node 4, and node 5, and the node 6 and the edge between node 6 and node 3 have been deleted.

[0055] Step B4: add 1 to the current edge degree as the current edge degree; repeat the above operation until there is no edge with a degree less than the current edge degree in the initial key edge set.

[0056] Specifically, after the edges with degrees less than the current edge degree are deleted from the question and answer sample library, 1 is added to the current edge degree as the current edge degree, the edges with degrees less than the current edge degree in the initial key edge set are deleted, and a new initial key edge set and a new (current edge degree minus 1) degree edge set are obtained. Repeat the above operation until there is no edge with a degree less than the current edge degree in the initial edge set.

[0057] For example, the initial edge parameter value is 2, and the current edge degree is 2. Assuming that the question and answer sample library includes edges with degrees of 1, 2, 3, 4, and 5, the edge with a degree of 1 is deleted from the question and answer sample library, and the remaining edge set with degrees of 2, 3, 4, and 5 is the initial key edge set. Further, 1 is added to the current edge degree to obtain a current edge degree of 3, the edges with degrees less than 3 in the initial key edge set are deleted, and a new initial key edge set is obtained as a set of edges with degrees of 3, 4, and 5. Repeat the above operation until the current edge degree is 6, the question and answer sample library only has a set of edges with a degree of 5, and after the edge with a degree of 5 is deleted, there is no edge with a degree less than 6 in the question and answer sample library. Stop updating, and obtain the final key edge set as the set of edges with a degree of 5.

[0058] Step B5: determining the anchor structure based on the key edge set.

[0059] The anchor structure is composed of key points and key edges. The key points are the points in the key edges. The structure composed of the key points and the key edges in the key edge set is the anchor structure. After obtaining the anchor structure, further, the position information of all nodes in the question and answer sample library based on the anchor structure is obtained.

[0060] In an optional implementation, the position information of each node can be determined according to the distance of each node to the key point in the anchor structure. For example, the smaller the node degree of a node, the greater the distance of the node to the anchor structure, and the more "remote" the position of the node in the question and answer sample library. The greater the node degree of a node, the smaller the distance of the node to the anchor structure, and the more "central" the position of the node in the question and answer sample library.

[0061] Through the above steps, the key points and key edges in the sample data can be determined before the sample information is input into the initial graph neural network model, and then the anchor structure and the position information of each point are determined according to the key edges. The problem of uneven data distribution is solved, and the training effect of the non-uniform graph neural network model is improved.

[0062] Step 202, extracting a sample from the question and answer sample library as a current sample, and constructing a corresponding graph data structure based on the current sample library.

[0063] The graph data structure includes a topological graph structure composed of nodes and edges, and position information of each node in the topological graph structure. In an optional implementation, during the training of the initial graph neural network, the encoder in the graph neural network model can encode the position information of the node in the topological graph into the node feature vector and serve as the input of the next encoding layer. Specifically, after determining the position information of each node in the topological graph composed of the question and answer sample library, the nodes, edges and position information of each node in the current sample are constructed into a graph data structure.

[0064] Step 203, inputting the graph data structure into the initial graph neural network model to obtain output information of the initial graph neural network model.

[0065] The output information is information corresponding to the input information and calculated by the initial graph neural network model. In an optional implementation, after the graph data structure constructed by the nodes, edges and position information of each node in the current sample is input into the initial graph neural network model, the initial graph neural network model can output output information in a certain format after calculation.

[0066] In step 204, the loss function of the initial graph neural network model is determined based on the output information of the initial graph neural network model and the sample label, and the network parameters in the initial graph neural network model are adjusted based on the loss function.

[0067] The sample label is the output information in the ideal state corresponding to the input information. The loss function is a function that maps the value of a random event or a random variable related to it to a non-negative real number to represent the "risk" or "loss" of the random event. The loss function is usually associated with optimization problems as a learning criterion, that is, the model is solved and evaluated by minimizing the loss function. For example, it is used for parameter estimation of the model in statistics and machine learning.

[0068] In practical applications, there is a certain "gap" between the output information obtained by the initial graph neural network model based on the input information and the sample label. The "gap" can be calculated by the loss function. Further, the network parameters in the initial graph neural network are adjusted according to the calculation result of the loss function until the initial graph neural network model meets the preset convergence condition and the adjustment is stopped. The adjusted initial graph neural network model is used as the trained non-uniform graph neural network model.

[0069] The training method of the non-uniform graph neural network model provided by the embodiment of the application can determine the question and answer sample library of the initial graph neural network and the sample label corresponding to the question and answer sample library, extract a sample from the question and answer sample library as a current sample, construct a corresponding graph data structure based on the current sample library, input the graph data structure into the initial graph neural network model to obtain the output information of the initial graph neural network, determine the loss function of the initial graph neural network model based on the output information of the initial graph neural network model and the sample label, and adjust the network parameters in the initial graph neural network based on the loss function. The technical solution of the embodiment can conveniently and quickly train the initial graph neural network model, and the position information of each point is concerned during the training process, the problem of uneven data distribution is solved, and the training effect of the non-uniform graph neural network model is improved. The trained non-uniform graph neural network model can analyze the problem information of the user, accurately and quickly output the answer information corresponding to the problem, process the problem quickly, and the accuracy of the output answer is high, further improving the user experience.

[0070] Figure 5 FIG. 1 is a structural schematic diagram of an intelligent question and answer implementation device provided by an embodiment of the application. The embodiment of the application provides an intelligent question and answer implementation device, which comprises:

[0071] The acquisition module 501 is configured to acquire the problem information input by the user and convert the problem information into target information.

[0072] The input module 502 is configured to input the target information into the pre-trained non-uniform graph neural network model to obtain output information of the non-uniform graph neural network model.

[0073] The output module 503 is configured to convert the output information into an answer corresponding to the question information and display the answer to the user.

[0074] Optionally, before obtaining the question information input by the user, the input module 502 is specifically configured to: if the non-uniform graph neural network model does not satisfy a pre-set convergence condition, extract a sample from a question and answer sample library as a current sample; the question and answer sample library includes nodes, edges, a pre-determined anchor structure, and position information of the nodes in the question and answer sample library.

[0075] The non-uniform graph neural network model is trained using the current sample until the non-uniform graph neural network model satisfies the convergence condition.

[0076] Optionally, the input module 502 is further configured to: determine the current sample and a sample label corresponding to the current sample.

[0077] Based on the current sample and the initial graph neural network model, output information of the initial graph neural network model is obtained.

[0078] Based on the output information of the initial graph neural network model and the sample label, a loss function of the initial graph neural network model is determined, and network parameters in the initial graph neural network model are adjusted based on the loss function until the non-uniform graph neural network model satisfies the convergence condition.

[0079] Optionally, the input module 502 is further configured to: construct a corresponding graph data structure based on the current sample.

[0080] The graph data structure is input into the initial graph neural network model to obtain the output information of the initial graph neural network model.

[0081] Optionally, the input module 502 is further configured to: obtain node degrees of all nodes in the question and answer sample library.

[0082] Based on the node degrees of all nodes in the question and answer sample library, edge degrees of all edges in the question and answer sample library are determined.

[0083] Based on the edge degrees of all edges in the question and answer sample library, a key edge set is determined, and the anchor structure is determined based on the key edge set.

[0084] Optionally, the input module 502 is further configured to: add or subtract 1 from the node degrees of two nodes connected by each edge to obtain the edge degree of each edge.

[0085] Optionally, the input module 502 is further configured to: take the initial edge parameter value as a current edge degree; delete edges with an edge degree less than the current edge degree in the question and answer sample library to obtain an initial key edge set and a current edge degree minus 1 degree edge set;

[0086] take the current edge degree plus 1 as the current edge degree; and repeat the above operations until there is no edge with an edge degree less than the current edge degree in the initial edge set.

[0087] The intelligent question and answer implementation device provided in the embodiments of the present application can execute the intelligent question and answer implementation method provided in any of the embodiments of the present application, and has the corresponding function modules and beneficial effects of the execution method.

[0088] Figure 6 is a structural schematic diagram of an electronic device provided by the embodiments of the present application, referring to Figure 6 which shows a structural schematic diagram of a computer system 12 of an electronic device suitable for implementing the embodiments of the present application. Figure 6 The electronic device shown is merely an example, and should not bring any limitation to the functions and use range of the embodiments of the present application. The components of the electronic device 12 can include but are not limited to one or more processors or processing units 16, a system memory 28, and a bus 18 connecting different system components including the system memory 28 and the processing unit 16.

[0089] The bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor or local bus using any of a variety of bus architectures. By way of example, these architectures include Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MAC) bus, Enhanced ISA bus, Video Electronics Standards Association (VESA) local bus, and Peripheral Component Interconnect (PCI) bus.

[0090] The electronic device 12 typically includes a variety of computer system readable media. These media can be any available media that is accessible by the electronic device 12 and includes both volatile and non-volatile media, removable and non-removable media.

[0091] The system memory 28 can include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The electronic device 12 can further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, a storage system 34 can be provided for reading from and writing to non-removable, non-volatile magnetic media (e.g., a "hard drive"). Figure 6 (not shown), commonly referred to as a "hard disk drive", for reading from and writing to non-removable, non-volatile magnetic media (e.g., a "hard drive"). Although not specifically shown, such computer system can further include other removable / non-removable, volatile / non-volatile computer system storage media including, but not limited to, magnetic cassette, memory cards or sticks, DVD, CDs, digital video tapes, solid state RAM, ROM, and the like. Figure 6A disk drive, a floppy disk drive, and / or other mass storage device can be provided for in the embodiment, and can be included in or peripheral to electronic device 12, as is known. A disk drive can be provided for reading from or writing to a removable, nonvolatile magnetic disk (e.g., "floppy disk"), and an optical disk drive can be provided for reading from or writing to a removable, nonvolatile optical disk (e.g., CD-ROM, DVD-ROM, etc.). In these instances, each drive can be connected to bus 18 by one or more data media interfaces. The drives can be external or internal to electronic device 12. Storage 28 can include one or more program products, which are configured to carry out the functions of embodiments of the application by embodying them in program modules 42 that execute on electronic device 12. Program / utility 40, having a set (e.g., at least one) of program modules 42, can be stored in storage by way of example, and not limitation, as well as an operating system, one or more application programs, other program modules, and program data, each or some combination thereof, can include implementation of a networking environment. Program modules 42 generally carry out the functions and / or methodologies of embodiments of the application as described herein.

[0092] Program / utility 40, having a set (at least one) of program modules 42, can be stored in storage by way of example, and not limitation, as well as an operating system, one or more application programs, other program modules, and program data, each or some combination thereof, can include implementation of a networking environment. Program modules 42 generally carry out the functions and / or methodologies of embodiments of the application as described herein.

[0093] Electronic device 12 can also communicate with one or more external devices 14 such as a keyboard or a pointing device, displays 24, etc.; one or more devices that enable a user to interact with electronic device 12; and / or one or more devices that enable electronic device 12 to communicate with one or more other computing devices. Such communication can occur via input / output (I / O) interface 22. Still yet, electronic device 12 in embodiments can communicate with one or more networks, such as one or more local area networks (LANs), wide area networks (WANs), and / or the Internet, through network adapter 20. As an example, network adapter 20 can include a modem, a network card (wireless or wired), or other well-known interface devices. Wireless networks can include cellular networks, Bluetooth, Wi-Fi, and / or the like. Wired networks can include the Internet, LAN, wide area networks (WAN), and / or the like. Wireless and / or wired networking interfaces can be provided by electronic device 12, and / or by display 24. As depicted, network adapter 20 communicates with the other components of electronic device 12 via bus 18. It should be appreciated that although not shown, other hardware and / or software modules can be used in conjunction with electronic device 12. Such hardware would include, but is not limited to, a microcode, a device driver, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archival storage systems, etc. Figure 6 Other hardware and / or software modules can be used in conjunction with electronic device 12 in embodiments, including, but not limited to, a microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archival storage systems, etc.

[0094] The processing unit 16 executes various functional applications and intelligent question answering implementations by running programs stored in the system memory 28, such as an intelligent question answering implementation method provided by the embodiments of the present application: obtaining question information input by a user, and converting the question information into target information; inputting the target information into a pre-trained heterogeneous graph neural network model to obtain output information of the heterogeneous graph neural network model; converting the output information into an answer corresponding to the question information, and displaying the answer to the user.

[0095] The embodiments of the present application provide a computer readable storage medium having a computer program stored thereon, which, when executed by a processor, implements an intelligent question answering implementation method provided by all embodiments of the present application: obtaining question information input by a user, and converting the question information into target information; inputting the target information into a pre-trained heterogeneous graph neural network model to obtain output information of the heterogeneous graph neural network model; converting the output information into an answer corresponding to the question information, and displaying the answer to the user. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium may, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination of the above. More specific examples (non-exhaustive list) of the computer readable storage medium include an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, the computer readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus or device.

[0096] The computer readable signal medium can include a data signal propagated in baseband or propagated as a carrier wave in a propagated data signal, in which the computer readable program code is embodied. Such propagated data signal can take a variety of forms, including but not limited to electro-magnetic, optical or any suitable combination thereof. The computer readable signal medium can also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate or transport program for use by or in connection with an instruction execution system, apparatus or device.

[0097] The program code contained in the computer readable medium can be transmitted by any suitable medium, including but not limited to wireless, wire, cable, optical fiber, RF, etc., or any suitable combination thereof.

[0098] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0099] It should be noted that the above-mentioned embodiments illustrate rather than limit the application, and that those skilled in the art will be able to design many alternative embodiments without departing from the scope of the application. The word "comprising" does not exclude the presence of elements or steps other than those listed and the word "a" or "an" preceding the usage of an element does not exclude the presence of a plurality of such elements. It is anticipated that, depending on the specific design choices made for a given embodiment, the application can be implemented in a variety of ways, all of which are intended to fall within the scope of the present application as defined by the subjoined claims.

Claims

1. An intelligent question-answering implementation method, characterized by, The method comprises: If the non-uniform graph neural network model does not satisfy the preset convergence condition, a sample in the question and answer sample library is extracted as a current sample; wherein the question and answer sample library comprises nodes, edges, a pre-determined anchor structure and position information of the nodes in the question and answer sample library; Node degrees of all nodes in the question and answer sample library are obtained; node degrees of two nodes connected by each edge are added and subtracted by 1 to obtain an edge degree of each edge; An initial edge parameter value is taken as a current edge degree; edges with an edge degree less than the current edge degree in the question and answer sample library are deleted to obtain an initial key edge set; the current edge degree is taken as the current edge degree; the above operations are repeatedly performed until there is no edge with an edge degree less than the current edge degree in the initial key edge set, a key edge set is obtained, and the anchor structure is determined based on the key edge set; The non-uniform graph neural network model is trained using the current sample until the non-uniform graph neural network model satisfies the convergence condition; wherein the non-uniform graph neural network model is used to process unstructured data, and the unstructured data refers to data of a problem with non-uniform distribution; Question information input by a user is obtained, and the question information is converted into target information; The target information is input into a pre-trained non-uniform graph neural network model to obtain output information of the non-uniform graph neural network model; The output information is converted into an answer corresponding to the question information, and the answer is displayed to the user.

2. The method of claim 1, wherein, The training of the non-uniform graph neural network model using the current sample until the non-uniform graph neural network model satisfies the convergence condition comprises: A sample label corresponding to the current sample is determined; Output information of the initial graph neural network model is obtained based on the current sample and the initial graph neural network model; A loss function of the initial graph neural network model is determined based on the output information of the initial graph neural network model and the sample label, and network parameters in the initial graph neural network model are adjusted based on the loss function until the non-uniform graph neural network model satisfies the convergence condition.

3. The method of claim 2, wherein, The obtaining of the output information of the initial graph neural network model based on the current sample and the initial graph neural network model comprises: A corresponding graph data structure is constructed based on the current sample; The graph data structure is input into the initial graph neural network model to obtain the output information of the initial graph neural network model.

4. An intelligent question-answering implementation apparatus characterized by comprising: The device comprises: An obtaining module is configured to extract a sample in a question and answer sample library as a current sample if a non-uniform graph neural network model does not satisfy a preset convergence condition; wherein the question and answer sample library comprises nodes, edges, a pre-determined anchor structure and position information of the nodes in the question and answer sample library; Node degrees of all nodes in the question and answer sample library are obtained; node degrees of two nodes connected by each edge are added and subtracted by 1 to obtain an edge degree of each edge; set the initial edge parameter value as a current edge degree; delete edges with an edge degree less than the current edge degree in the question and answer sample library to obtain an initial key edge set; set the current edge degree as 1 more than the current edge degree; repeat the above operations until there is no edge with an edge degree less than the current edge degree in the initial key edge set to obtain a key edge set, and determine the anchor structure based on the key edge set; training the non-uniform graph neural network model using the current sample until the non-uniform graph neural network model meets the convergence condition; wherein the non-uniform graph neural network model is used to process unstructured data, and the unstructured data refers to data of a problem with non-uniform distribution; obtaining question information input by a user, and converting the question information into target information; an input module configured to input the target information into a pre-trained non-uniform graph neural network model to obtain output information of the non-uniform graph neural network model; an output module configured to convert the output information into an answer corresponding to the question information, and display the answer to the user.

5. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the intelligent question and answer implementation method of any one of claims 1-3.

6. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the intelligent question and answer implementation method of any one of claims 1-3.

Images