CN115730047A - Intelligent question-answering method, equipment, device and storage medium - Google Patents

Abstract

The application provides an intelligent question answering method, equipment, a device and a storage medium; in the method, a target question provided by a target object is obtained, and after a target answer corresponding to the target question is not found in a question-answer knowledge base storing question-answer relationship pairs, feature vector extraction is performed on each participle in the target question based on context statement information corresponding to each participle in the target question to obtain corresponding first semantic features; searching a standard question with the similarity reaching a similarity threshold value with the semantic feature similarity between the question and the target question in a preset question-answering library based on the first semantic feature; acquiring a first standard answer corresponding to the standard question, and taking the first standard answer as a target answer corresponding to the target question; the first semantic features corresponding to the target problem are extracted by combining the whole statement information of the target problem instead of only referring to a single keyword, so that the identification accuracy is improved, the target answer can be directly returned, and the search efficiency is improved.

Description

Intelligent question and answer method, equipment, device and storage medium

Technical Field

The present application relates to the field of natural language processing technologies, and in particular, to an intelligent question answering method, device, apparatus, and storage medium.

Background

With the development of artificial intelligence technology, the intelligent question-answering technology has the advantage of saving human resources, and is widely applied to various industries.

At present, a keyword plus sequencing mode is adopted in an intelligent question-answering technology to identify, match and feed back problems proposed by a target object. Specifically, a keyword mode is adopted to identify and match the questions proposed by the target object, a plurality of possible questions identified and matched are ranked and fed back to the target object, the target object screens the possible questions, a specific transaction link corresponding to the screened questions of the target object is finally returned, and the target object clicks the link to obtain the answer to the question.

Obviously, the search requirement of the target object can be met through the current mode, but specific answers cannot be directly given, and when the recognition matching is carried out in a keyword adding and sorting mode, the recognition matching cannot be effectively carried out on fine-grained problems, so that the answers cannot be accurately matched.

Therefore, how to accurately identify the question posed by matching with the target object to accurately match the answer and improve the search efficiency is a problem that needs to be solved at present.

Disclosure of Invention

The application provides an intelligent question-answering method, equipment, a device and a storage medium, which are used for improving the search efficiency and the accuracy of answers matched with target questions.

In a first aspect, an embodiment of the present application provides an intelligent question answering method, including:

acquiring a target problem proposed by a target object;

after a target answer corresponding to the target question is not found in a question-answer knowledge base storing the question-answer relationship pairs, extracting feature vectors of all the participles in the target question based on context statement information corresponding to all the participles in the target question to obtain corresponding first semantic features;

searching a standard question with the semantic similarity reaching a similarity threshold value with the target question in a preset question-answering library based on the first semantic features;

and acquiring a first standard answer corresponding to the standard question, and taking the first standard answer as a target answer corresponding to the target question.

In the embodiment of the application, a target question provided by a target object is obtained, and after a target answer corresponding to the target question is not found in a question-answer knowledge base storing question-answer relationship pairs, feature vector extraction is carried out on each word segmentation in the target question to obtain corresponding first semantic features; based on the first semantic features, searching a standard question with the semantic similarity reaching a similarity threshold value with the target question in a preset question-answer base, and based on the overall semantic matching of the target question and the standard question with the similarity of the target question, improving the accuracy of the target question matching; and a first standard answer corresponding to the standard question is obtained, the first standard answer is used as a target answer corresponding to the target question, the answer corresponding to the target question is directly fed back to the target object, the answer is not required to be obtained through a link, and the searching efficiency is improved.

In one possible implementation manner, the target answer corresponding to the target question is searched in the question-answer knowledge base by the following method:

performing word segmentation processing on the target problem, and determining key words and part-of-speech information of the key words in the target problem;

matching the target question with a question template in a question and answer knowledge base based on the keyword and the part-of-speech information, and obtaining a matching result;

and searching a target answer corresponding to the target question in a question-answer knowledge base based on the matching result.

In a possible implementation manner, based on the matching result, searching a question-answer knowledge base for a target answer corresponding to a target question specifically includes:

if the matching result represents that the target question template matched with the target question exists, acquiring a second standard answer corresponding to the target question template in a question-answer knowledge base, and taking the second standard answer as a target answer corresponding to the target question; or

And if the matching result represents that the target question template matched with the target question does not exist, determining that the target answer corresponding to the target question is not found in the question-answer knowledge base.

In the embodiment of the application, the question template matched with the target is matched in the question-answer knowledge base storing the question-answer relationship pairs, and the target answer corresponding to the target question is directly returned after the matching is successful, so that the search efficiency and the accuracy of the answer are improved.

In a possible implementation manner, after acquiring a target question posed by a target object, before searching a target answer corresponding to the target question in a question-answer knowledge base storing a question-answer relationship pair, the method further includes:

identifying the item type corresponding to the target problem through a text convolution neural network;

searching a question set corresponding to the question type in a question-answer knowledge base;

and searching the item problem matched with the target problem in the item problem set.

In the embodiment of the application, before searching the target answer corresponding to the target question in the question-answer knowledge base storing the question-answer relationship pair, the item type corresponding to the target question is identified, the item problem set corresponding to the item type is searched in the question-answer knowledge base, and then the item problem matched with the target question is searched in the item problem set, so that the complexity of matching the target question can be reduced, and the accuracy of matching the target question can be improved.

In a possible implementation manner, extracting a feature vector for each participle in the target problem to obtain a corresponding first semantic feature includes:

extracting feature vectors of all the participles in the target problem through an ALBert model to obtain corresponding first semantic features;

based on the first semantic features, searching a standard question with the semantic similarity reaching a similarity threshold value with the target question in a preset question-answering library, wherein the standard question comprises the following steps:

aiming at any preset problem in a preset question-answering library, determining semantic similarity between the preset problem and a target problem through an Euclidean distance algorithm based on a first semantic feature and a second semantic feature corresponding to the preset problem;

and searching the preset questions with the semantic similarity reaching the similarity threshold value from all the preset questions stored in the preset question-answer library based on the semantic similarity, and taking the preset questions obtained through searching as standard questions.

In the embodiment of the application, feature vectors are extracted from each participle in a target problem through an ALBert model, the feature vectors are constructed based on target problem information, and the codes of the same word in different target problems are different, namely, the acquired first semantic features are combined with the target problem and are not performed based on a single word; therefore, when the semantic similarity between the preset problem and the target problem is determined based on the first semantic feature and the second semantic feature corresponding to the preset problem, the semantic similarity is based on the semantics of the whole target problem and is not based on a single keyword, when the semantic similarity reaches the preset problem of the similarity threshold value, the searched preset problem is more consistent with the target problem, and the searching accuracy is improved.

In a possible implementation manner, acquiring a first standard answer corresponding to a standard question, and taking the first standard answer as a target answer of a target question specifically includes:

if the number of the obtained standard problems is determined to exceed the number threshold, screening out the standard problems with the semantic similarity arrangement serial numbers larger than the set serial numbers from the obtained standard problems, and taking all first standard answers corresponding to the screened standard problems as target answers; or

If the number of the obtained standard questions is determined to exceed the number threshold, for any standard question, extracting a set number of first standard answers from first standard answers corresponding to the standard questions, and taking the extracted first standard answers as target answers.

In the embodiment of the application, when the number of the obtained standard questions exceeds a threshold value, each standard question corresponds to at least one first standard answer, a large number of first standard answers are obtained at the moment, if the first standard answers are all fed back to the target object, the target object cannot conveniently search the required target answers among the large number of first standard answers, therefore, in order to enable the target object to more quickly obtain the required target answers, the obtained standard questions are subjected to secondary screening, the first standard answers corresponding to the standard questions subjected to the secondary screening are returned to the target object, or the first standard answers corresponding to the standard questions are extracted, the extracted first standards are fed back to the answer target object, the number of the first standard answers is reduced, and the target object can conveniently obtain the required target answers.

In one possible implementation manner, if a standard question is not retrieved from a preset question-answer library based on a first semantic feature, recording the first semantic feature and the number of questions corresponding to the first semantic feature in a message log;

and when the questioning times reach a time threshold value, issuing the target questions corresponding to the first semantic features to update a preset questioning and answering library.

In the embodiment of the application, when the standard question is not retrieved from the preset question-answer library, the first semantic feature and the question number corresponding to the first semantic feature are recorded, when the question number reaches a number threshold, the target question corresponding to the first semantic feature is issued to update the preset question-answer library, the preset question-answer library is updated perfectly, the target answer corresponding to the target question is more conveniently obtained, the phenomenon that the target answer corresponding to the target question cannot be searched in a matching manner is reduced, and the use experience of the target object is improved.

In a second aspect, an embodiment of the present application provides an intelligent question answering device, including: human-computer interaction interface, memory and processor, wherein:

the man-machine interaction interface is used for acquiring a target problem proposed by a target object;

a memory for storing a computer program operable on the processor;

a processor for reading the computer program in the memory and executing: after a target answer corresponding to the target question is not found in a question-answer knowledge base storing the question-answer relationship pairs, extracting feature vectors of all the participles in the target question based on context statement information corresponding to all the participles in the target question to obtain corresponding first semantic features; searching a standard question with the semantic similarity reaching a similarity threshold value with the target question in a preset question-answer base based on the first semantic features; and acquiring a first standard answer corresponding to the standard question, and taking the first standard answer as a target answer corresponding to the target question.

In a third aspect, an embodiment of the present application provides an intelligent question answering device, including:

the first acquisition module is used for acquiring a target problem proposed by a target object;

the extraction module is used for extracting the feature vector of each participle in the target question based on the context statement information corresponding to each participle in the target question to obtain corresponding first semantic features after the target answer corresponding to the target question is not found in the question-answer knowledge base storing the question-answer relation pair;

the retrieval module is used for retrieving a standard question with the semantic similarity reaching a similarity threshold value with the target question from a preset question-answering library based on the first semantic features;

and the second acquisition module is used for acquiring a first standard answer corresponding to the standard question and taking the first standard answer as a target answer corresponding to the target question.

In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, where computer instructions are stored, and when the computer instructions are executed by a processor, the intelligent question answering method provided in the embodiment of the present application is implemented.

Additional features and advantages of the application will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by the practice of the application. The objectives and other advantages of the application may be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.

Fig. 1 is a schematic view of an application scenario provided in an embodiment of the present application;

fig. 2 is a schematic structural diagram of an intelligent question answering device according to an embodiment of the present application;

fig. 3 is a flowchart of a method for intelligent question answering according to an embodiment of the present application;

FIG. 4 is a schematic illustration of a knowledge-graph provided by an embodiment of the present application;

FIG. 5 is a flowchart of another method for intelligent question answering provided by an embodiment of the present application;

fig. 6 is a schematic structural diagram of an intelligent question answering device according to an embodiment of the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more clearly understood, the technical solutions in the embodiments of the present application will be described below clearly and completely with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only some embodiments of the present application, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

The terms "first," "second," and the like in the description and in the claims of the present application and in the above-described drawings are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the application described herein are capable of operation in sequences other than those illustrated or described herein.

The embodiment of the application relates to a natural language processing technology, is applied to an intelligent question answering process and is mainly used for answering questions which are provided by a target object in a natural language form; the intelligent question answering method provided by the embodiment of the application is widely applied to various scenes of life. The classic application scenario includes: intelligent voice interaction, online customer service, knowledge acquisition, emotional chatting, and the like. Common classifications are: generating a modeling and retrieval type question-answering system; single-round question-answering and multi-round question-answering systems; the question-answering system is oriented to the open field and the specific field.

And the intelligent question-answering process has different solutions to different types of problems. Generally, questions are classified into question-answer type, task type, chatting type, and the like. Wherein, for question-answer type questions, the target object hopes to get the answer of the question, at this moment, the answer to the question comes from the specific knowledge base, reply the target object with the specific answer; for the task type problem, the target object hopes to complete a specific task, and at the moment, the target object is helped to complete a specified task through the semantic execution background butt joint capability; for the chatty type question, the target object has no explicit purpose, and there is no second standard answer at this time. The intelligent question answering method provided by the embodiment of the application mainly aims at question answering type questions.

In the related technology, in the process of intelligent question answering, when answering questions in question answering mode, at least one question with high similarity to a target question proposed by a target object is obtained mainly in a keyword plus sequencing mode, the obtained at least one question is fed back to the target object in a sequencing mode, the fed-back question is screened by the target object, a transaction link corresponding to the screened question is obtained, and an answer can be obtained only after the transaction link is further clicked. Although the search function is realized, the process is complex, answers to questions cannot be directly fed back to the target object, the search efficiency is low, and the answers cannot be accurately matched due to the fact that the fine-grained questions cannot be effectively identified and matched by means of the keywords.

In view of the above, based on the problems in the related art, and considering that the complexity of the target object in handling government affair questions is usually high, and the content cannot be obtained by extracting keywords and matching the questions, the embodiment of the present application provides an intelligent question answering method;

in the embodiment of the application, a target question provided by a target object is obtained, and after a target answer corresponding to the target question is not found in a question-answer knowledge base storing question-answer relationship pairs, feature vector extraction is performed on each participle in the target question based on context statement information corresponding to each participle in the target question to obtain a corresponding first semantic feature; searching a standard question with the semantic similarity reaching a similarity threshold value with the target question in a preset question-answer base based on the first semantic features; and acquiring a first standard answer corresponding to the standard question, and taking the first standard answer as a target answer corresponding to the target question.

In the implementation of the application, the question matched with the target question is searched in the question and answer knowledge base in which the question and answer relation pairs are stored, and the question and answer knowledge base stores the fixed question and answer relation pairs, so that after the question matched with the target question is searched in the question and answer knowledge base, the corresponding answer can be obtained, the target answer can be directly returned, and the search efficiency is improved; if the target answer corresponding to the target question is found in the question-answer knowledge base storing the question-answer relationship pairs, in order to ensure the accuracy of matching identification, in the embodiment of the application, feature vector extraction is performed on each participle in the target question based on context statement information corresponding to each participle in the target question to obtain corresponding first semantic features, that is, the first semantic features are extracted in combination with the overall semantics of the target question instead of only referring to a single keyword, so that the identification accuracy is improved, the target answer can be directly returned, and the search efficiency is improved.

After introducing the design concept of the embodiment of the present application, some simple descriptions are provided below for application scenarios to which the technical solution of the embodiment of the present application can be applied, and it should be noted that the application scenarios described below are only used for describing the embodiment of the present application and are not limited. In a specific implementation process, the technical scheme provided by the embodiment of the application can be flexibly applied according to actual needs.

Referring to FIG. 1, FIG. 1 schematically provides an application scenario of the embodiment of the present application, which includes an intelligent question answering device 10 (e.g., 10-1 and 10-2 shown in FIG. 1) and a server 20;

the intelligent question-answering device 10 can be a terminal device such as a personal computer, a mobile phone, a tablet computer, a notebook computer and a question-answering robot;

the server 20 may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing basic cloud computing services such as a cloud service, a cloud database, cloud computing, a cloud function, cloud storage, a Network service, cloud communication, a middleware service, a domain name service, a security service, a Content Delivery Network (CDN), a big data and artificial intelligence platform, and the like.

In one possible embodiment, the intelligent question answering device 10 and the server 20 may communicate with each other through a communication network, which may be a wired network or a wireless network. The smart question-answering device 10 and the server 20 may be directly or indirectly connected through wired or wireless communication. For example, the intelligent question answering device 10 may be indirectly connected with the server 20 through the wireless access point 11, or the intelligent question answering device 10 may be directly connected with the server 20 through the internet, which is not limited herein.

In a possible application scenario, the intelligent question-answering device 10 obtains a question-answer relationship pair and a preset question-answer library sent by the server 20, and stores information sent by the server 20; then, after the intelligent question-answering device 10 acquires the target question provided by the target object through the human-computer interaction interface, the target question is identified, a target answer corresponding to the target question is determined, and the target answer is displayed on the display unit of the intelligent question-answering device 10.

In another possible application scenario, after obtaining a target question posed by a target object, the smart question-answering device 10 sends the target question to the server 20, the server 20 executes a process of determining a target answer corresponding to the target question, and feeds the determined target answer back to the smart question-answering device 10 for display by the smart question-answering device 10.

Referring to fig. 2, fig. 2 exemplarily provides a schematic structural diagram of an intelligent question answering device 10 in the embodiment of the present application.

It should be understood that the smart question-answering device 10 shown in fig. 2 is only one example, and that the smart question-answering device 10 may have more or less components than those shown in fig. 2, may combine two or more components, or may have a different configuration of components. The various components shown in fig. 2 may be implemented in hardware, software, or a combination of hardware and software, including one or more signal processing and/or application specific integrated circuits.

A block diagram of a hardware configuration of the intelligent question answering device 10 according to an exemplary embodiment is exemplarily shown in fig. 2. As shown in fig. 2, the intelligent question-answering apparatus 10 includes: radio Frequency (RF) circuit 100, memory 101, human-computer interaction interface 102, display unit 103, camera 104, communication interface 105, wireless Fidelity (Wi-Fi) module 106, processor 107, power supply 108, and the like.

The RF circuit 100 may be used for receiving and transmitting signals during information transmission and reception or during a call, and may receive downlink data of a base station and then send the downlink data to the processor 107 for processing; the uplink data may be transmitted to the base station. Typically, the RF circuitry includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low noise amplifier, a duplexer, and the like.

The memory 101 is used to store computer programs and data that can be run on the processor 107; the computer program, when executed by the processor 107, causes the processor 107 to perform each step in the intelligent question-answering method of the various exemplary embodiments of this application.

In one possible implementation, the memory 101 may include a readable medium in the form of volatile memory, such as high-speed Random Access Memory (RAM) and/or cache memory, and may further include Read Only Memory (ROM); non-volatile memory may also be included, such as at least one magnetic disk storage device, flash memory device, or other volatile solid state storage device.

Memory 101 may also include a program/utility having a set (at least each) of program modules, please refer to fig. 2, which is an exemplary partial block diagram of memory 101 provided in fig. 2; wherein the program modules include, but are not limited to: an operating system, each or more application programs, other program modules, and program data.

In this embodiment, the memory 101 may further store a question-answer relationship pair in the question-answer knowledge base, and a preset standard question in the question-answer knowledge base and a first standard answer corresponding to the standard question.

The human-computer interaction interface 102 may be configured to receive input numeric or character information and generate signal inputs related to user settings and function control of the intelligent question-answering device 10, and specifically, the input interface 120 may include a touch panel 1021 arranged on the front of the intelligent question-answering device 10 for collecting touch operations of a user on the touch panel 1021 and determining various instructions triggered by the user, such as clicking a button, dragging a scroll box, and the like. For example, in the embodiment of the present application, the target object inputs the target question to be asked through the human-computer interaction interface 102, that is, the intelligent question-answering device 10 obtains the target question posed by the target object through the human-computer interaction interface 102.

In one possible implementation, the human-computer interaction interface 102 further includes other input devices 1022, such as voice input, etc., for recognizing various instructions triggered by the user by receiving a voice signal sent by the user.

The display unit 103 may be used to display information input by the user or information provided to the user and a Graphical User Interface (GUI) of various menus of the intelligent question-answering device 10. Specifically, the display unit 103 may include a display panel 1031 provided on the front surface of the smart question-answering device 10. The display panel 1031 may be configured in the form of a liquid crystal display, a light emitting diode, or the like. The display unit 103 may be configured to display a target answer corresponding to a target question in the embodiment of the present application.

In one possible implementation manner, the touch panel 1021 may be covered on the display panel 1031, or the touch panel 1021 and the display panel 1031 may be integrated to implement the input and output functions of the intelligent question answering device 10, and after the integration, the integration may be referred to as a touch display screen for short. In the present application, the display unit 103 may display the application programs and the corresponding operation steps.

The camera 104 may be used to capture still images or video. The object generates an optical image through the lens and projects the optical image to the photosensitive element. The photosensitive element may be a Charge Coupled Device (CCD) or a complementary metal-oxide-semiconductor (CMOS) phototransistor. The light sensitive elements convert the optical signals into electrical signals which are then passed to a processor 107 for conversion into digital image signals.

The communication interface 105 is used to transmit various instructions or data acquired by the intelligent question answering device 10 to the server 20, and to acquire instructions or data from the server 20 through the communication interface 105.

Wi-Fi belongs to a short-distance wireless transmission technology, and the intelligent question-answering device 10 can help a user to receive and send emails, browse webpages, access streaming media and the like through the Wi-Fi module 106, and provides wireless broadband internet access for the user.

The processor 107 is a control center of the intelligent question answering apparatus 10, connects the respective parts of the entire intelligent question answering apparatus 10 by various interfaces and lines, performs various functions of the intelligent question answering apparatus 10 and processes data by running or executing software programs for intelligent question answering stored in the memory 101, and calling data stored in the memory 101.

In some embodiments, processor 107 may include one or more processing units; the processor 107 may also integrate an application processor, which primarily handles operating systems, user interfaces, applications, etc., and a baseband processor, which primarily handles wireless communications. It will be appreciated that the baseband processor described above may not be integrated into the processor 107. In the embodiment of the present application, the processor 107 may run an operating system and a response message, and the intelligent question answering method in the embodiment of the present application.

In this embodiment of the application, the processor 107 is configured to, after a target answer corresponding to a target question is not found in a question-answer knowledge base storing a question-answer relationship pair, perform feature vector extraction on each participle in the target question based on context statement information corresponding to each participle in the target question, and obtain a corresponding first semantic feature; searching a standard question with the semantic similarity reaching a similarity threshold value with the target question in a preset question-answering library based on the first semantic features; and acquiring a first standard answer corresponding to the standard question, and taking the first standard answer as a target answer corresponding to the target question.

In one possible implementation, the processor 107 is further configured to search the question-answer knowledge base for a target answer corresponding to the target question by:

performing word segmentation processing on the target problem, and determining key words and key part-of-speech information in the target problem;

matching the target question with a question template in a question-answer knowledge base based on the keyword and the part-of-speech information, and obtaining a matching result;

and searching a target answer corresponding to the target question in a question-answer knowledge base based on the matching result.

In one possible implementation, the processor 107 is specifically configured to:

if the matching result represents that the target question template matched with the target question exists, acquiring a second standard answer corresponding to the target question template in a question-answer knowledge base, and taking the second standard answer as a target answer corresponding to the target question; or

And if the matching result represents that the target question template matched with the target question does not exist, determining that the target answer corresponding to the target question is not found in the question-answer knowledge base.

In a possible implementation manner, after obtaining the target question posed by the target object, before looking up the target answer corresponding to the target question in the question-answer knowledge base storing the question-answer relationship pair, the processor 107 is further configured to:

identifying the item type corresponding to the target problem through a text convolution neural network;

searching a question set corresponding to the question type in a question-answer knowledge base;

and searching the item problem matched with the target problem in the item problem set.

In one possible implementation, the processor 107 is specifically configured to:

extracting feature vectors of all the participles in the target problem through an ALBert model to obtain corresponding first semantic features;

aiming at any preset problem in a preset question-answering library, determining semantic similarity between the preset problem and a target problem through an Euclidean distance algorithm based on a first semantic feature and a second semantic feature corresponding to the preset problem;

and searching the preset questions with the semantic similarity reaching the similarity threshold value from all the preset questions stored in the preset question-answer library based on the semantic similarity, and taking the preset questions obtained through searching as standard questions.

In one possible implementation, the processor 107 is specifically configured to:

if the number of the obtained standard problems is determined to exceed the number threshold, screening out the standard problems with the semantic similarity arrangement serial numbers larger than the set serial numbers from the obtained standard problems, and taking all first standard answers corresponding to the standard problems as target answers; or

If the number of the obtained standard questions is determined to exceed the number threshold, for any standard question, extracting a set number of first standard answers from first standard answers corresponding to the standard questions, and taking the extracted first standard answers as target answers.

In one possible implementation, the processor 107 is further configured to:

if the standard question is not retrieved in the preset question-answering base based on the first semantic feature, recording the first semantic feature and the question number corresponding to the first semantic feature in the message log;

and when the number of questioning times reaches a time threshold value, issuing the target question corresponding to the first semantic feature to update a preset question-answer library.

It should be noted that the intelligent question answering device 10 provided in the embodiment of the present application further includes a bluetooth module, a sensor, an audio circuit, a speaker, a microphone, and the like. Wherein:

and the Bluetooth module is used for performing information interaction with other Bluetooth equipment with the Bluetooth module through a Bluetooth protocol. For example, the smart question-answering device 10 can establish a bluetooth connection with a wearable electronic device (e.g., a smart watch) that is also equipped with a bluetooth module through the bluetooth module, so as to perform data interaction.

The smart question-answering device 10 may also include at least one sensor, such as an acceleration sensor, a distance sensor, a fingerprint sensor, a temperature sensor. The intelligent question-answering device 10 may also be equipped with other sensors such as gyroscopes, barometers, hygrometers, thermometers, infrared sensors, light sensors, motion sensors, and the like.

The audio circuitry, speaker, microphone may provide an audio interface between the user and the intelligent question-answering device 10. The audio circuit can transmit the electric signal converted from the received audio data to the loudspeaker, and the electric signal is converted into a sound signal by the loudspeaker to be output. The smart question-answering device 10 may also be provided with a volume button for adjusting the volume of the sound signal. On the other hand, a microphone converts a collected sound signal into an electrical signal, which is received by an audio circuit and converted into audio data, which is then output to an RF circuit for transmission to, for example, another terminal, or to a memory for further processing. The microphone in the application can acquire the voice of a user.

The smart question-answering device 10 also includes a power source 108 (such as a battery) for powering the various components. The power supply may be logically coupled to the processor 107 through a power management system to manage charging, discharging, and power consumption functions through the power management system. The intelligent question-answering device 10 may also be configured with power buttons for powering on and off the terminal, and locking the screen.

The intelligent question answering method provided by the exemplary embodiment of the present application is described below with reference to the accompanying drawings in conjunction with the application scenarios described above, it should be noted that the application scenarios described above are only shown for the convenience of understanding the spirit and principles of the present application, and the embodiments of the present application are not limited in this respect.

Referring to fig. 3, fig. 3 exemplarily provides an intelligent question answering method in an embodiment of the present application, where the method includes:

step S300, a target question proposed by a target object is acquired.

Step S301, after a target answer corresponding to the target question is not found in a question-answer knowledge base storing question-answer relationship pairs, extracting feature vectors of all the participles in the target question based on context statement information corresponding to all the participles in the target question to obtain corresponding first semantic features.

When a target question is obtained, in order to quickly and accurately obtain a target answer corresponding to the target question, in the embodiment of the application, a question-answer knowledge base storing question-answer relation pairs is firstly searched for a target question template matched with the target question, and a matching result is determined; and determining whether a target answer corresponding to the target question is found from the question-answer knowledge base or not based on the matching result.

In one possible implementation, the question-answer relationship pairs in the question-answer knowledge base are stored in the form of a knowledge graph, please refer to fig. 4, where fig. 4 exemplarily provides a schematic diagram of the knowledge graph;

it should be noted that different fields correspond to different knowledge maps, so that corresponding knowledge maps are required to be separately established for different fields to form a question-answer knowledge base; for example, when the embodiment of the application is used for intelligently answering government affairs, the entity and the relation need to be designed according to matters in government affairs, and a government affair knowledge map is built to form a government affair question and answer knowledge base.

When a target answer corresponding to a target question is searched in a question and answer knowledge base, matching the target question with a question template in the question and answer knowledge base to obtain a matching result, and then searching the target answer corresponding to the target question in the question and answer knowledge base based on the matching result;

because the target question is matched with the question template in the question and answer knowledge base, in the embodiment of the application, a keyword slot and a query statement are configured in advance for the relation defined in the question and answer knowledge base according to the data of the question and answer items and the possible questions to form a corresponding question template;

therefore, when the target question is matched with the question template in the question-answer knowledge base, word segmentation processing is firstly carried out on the target question to determine the keywords in the target question and the part-of-speech information of the keywords; it should be noted that, in the process of performing word segmentation processing, special processing is performed on special proper nouns in the field according to the keyword dictionary of the corresponding field in addition to the natural language processing tool;

after the keywords and the part-of-speech information of the keywords are obtained, matching the keywords with a question template in a question and answer knowledge base based on the part-of-speech information of the keywords, namely, matching the keywords with a keyword slot in the question template to obtain a matching result;

if the matching result represents that the target problem template matched with the target problem is found in the question and answer knowledge base, namely the matching is successful, a problem which corresponds to the target problem and needs to be solved is obtained at the moment, the problem which needs to be solved and the query sentence are searched in the question and answer knowledge base based on the problem which needs to be solved and the second standard answer corresponding to the problem which needs to be solved can be found, and the second standard answer is used as the target answer corresponding to the target problem to be fed back;

and if the matching result represents that the target question template matched with the target question is not found in the question-answer knowledge base, determining that the target answer corresponding to the target question is not found in the question-answer knowledge base storing the question-answer relationship pair.

In the embodiment of the application, after it is determined that the target answer corresponding to the target question is not found in the question and answer knowledge base, a standard question similar to the target question is obtained in a deep learning mode, a first standard answer corresponding to the standard question is obtained, and the first standard answer is used as the answer of the target question.

In a possible implementation mode, in the process of acquiring a standard problem similar to a target problem by adopting a deep learning mode, firstly performing word segmentation processing on the target problem, and then performing feature vector extraction on each word segmentation of the target problem based on context statement information of each word segmentation in the target problem to obtain corresponding first semantic features;

performing word segmentation Processing on the target problem based on a pre-constructed and maintained noun dictionary and a Natural Language Processing (NLP) tool to obtain each word segmentation in the target problem, and identifying keywords in the target problem; it should be noted that different noun dictionaries correspond to different fields, for example, when the embodiment of the present application performs intelligent question answering on government affair questions, the noun dictionary is a government affair special noun dictionary which is constructed and maintained in advance;

after each participle of the target problem is obtained, inputting each participle of the target problem into an ALBert model, and performing feature vector extraction on each participle of the target problem through the ALBert model to obtain first semantic features corresponding to each participle of the target problem.

Because deep Neural Network models such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) need to be input in a numerical form for words, the participles in the target problem are converted into feature vectors during encoding, and the meanings of the words are identified in the form of the feature vectors. A commonly used feature vector method is One Hot coding (One Hot), which causes data sparseness and dimensionality disaster and is generally not suitable when a dictionary is large; a coding method based on distributed expression, such as Word2Vec, adopts a method of constructing a neural network to learn Word vectors, but the coding is the same for the Word vectors of the same words, so that the problem of Word ambiguity cannot be solved; therefore, the method for extracting the feature vector by using the ALBert model and the coding method in the ALBert model is provided in the embodiment of the application, the word vector is constructed based on sentence information, the codes of the same words in different sentences are different, and semantic features can be extracted well.

After the first semantic features corresponding to the target questions are extracted, the standard questions similar to the target questions are searched in a preset question-answering base based on the first semantic features.

Step S302, based on the first semantic features, searching a standard question with the semantic similarity reaching a similarity threshold value with the target question in a preset question-answering library.

In the embodiment of the application, the preset problems in the preset question-answering library are the problems that the knowledge graph is difficult to build through a unified mode, and most of the problems are arranged in a manual mode. Taking intelligent question answering aiming at government affair questions as an example, the questions and the corresponding answers of the government affair matters are screened, and a preset question-answer library is determined based on the screening result.

In a possible implementation manner, the preset questions and corresponding answers in the preset question-and-answer library may be in a form of a table, please refer to table 1, where table 1 exemplarily provides a corresponding relationship between the questions and the answers in the preset question-and-answer library, the left column is the questions, and the right column is the answers;

TABLE 1

Based on semantic features, retrieving standard questions similar to the target questions from a preset question-answering library; that is to say, determining semantic similarity between a first semantic feature corresponding to the target problem and a second semantic feature corresponding to the preset problem, wherein the semantic similarity is used for representing the similarity between the target problem and the preset problem; therefore, it is necessary to determine corresponding second semantic features for each standard question stored in the preset question-answering database, and the determination manner of the second semantic features may be the same as that of the first semantic features, which is not described herein again.

In the embodiment of the present application, when determining semantic similarity between a target question and each preset question, for any preset question in a preset question-and-answer library, the following steps are performed:

determining semantic similarity between a preset problem and a target problem through an Euclidean distance algorithm based on the first semantic feature and the second semantic feature;

the Euclidean distance algorithm corresponds to the following formula:

wherein x and y respectively represent word vectors in the target question and the preset question.

Because the similarity is calculated by adopting algorithms such as BoW, TF-IDF, jaccord and the like, the literal similarity is mainly solved, but each participle in the target problem has rich meanings in the target problem, and the semantic similarity between two sentences cannot be determined directly according to keyword matching or a shallow model based on machine learning, the Euclidean distance algorithm is adopted in the embodiment of the application to improve the accuracy of the semantic similarity between the two sentences.

And after the semantic similarity between each preset problem and the target problem is obtained, searching the preset problems of which the semantic similarity reaches a similarity threshold value, and taking the preset problems obtained by searching as standard problems.

Step S303, a first standard answer corresponding to the standard question is obtained, and the first standard answer is used as a target answer corresponding to the target question.

In the embodiment of the application, in the process of acquiring the first standard answers corresponding to the standard problems, if the number of the standard problems acquired by searching is determined to exceed the number threshold, the standard problems with the semantic similarity arrangement sequence number larger than the set sequence number are screened out from the standard problems acquired by searching, and all the first standard answers corresponding to the standard problems screened out twice are used as target answers; or

If the number of the retrieved standard questions exceeds the number threshold, extracting a set number of first standard answers from all first standard answers corresponding to any standard question, and taking the extracted first standard answers as target answers.

In a possible implementation manner, a situation that a standard is not retrieved in a preset question and answer library exists, at this time, a behavior of a target object cannot be fed back and analyzed in time, and the target object cannot be extracted from a diversity problem of the target object. Therefore, in the embodiment of the application, based on the first semantic feature, if the standard question is not retrieved from the preset question-answering library, the first semantic feature is recorded in the message log, and the number of times of question asking corresponding to the first semantic feature is used; and when the number of questioning times reaches a time threshold value, issuing the target question corresponding to the first semantic feature to update a preset question-answer library.

In another possible implementation manner, in order to improve the efficiency and accuracy of searching, after a target problem proposed by a target object is obtained, intention identification is performed based on the target problem; because different fields are widely distributed, different matters are contained, such as intelligent questions and answers aiming at government affairs, and the government affairs field contains matters such as social security, insurance, accumulation fund and the like; meanwhile, the problems corresponding to each item are many, and if the intention is not identified, the problems are matched with all the problems in the field, so that the calculation amount is large, and the efficiency is low.

Illustratively, when the intention is identified, inputting the target problem into a text convolution neural network, and identifying the item type corresponding to the target problem through the text convolution neural network; and then searching a matter problem set corresponding to the matter type in a question and answer knowledge base, and searching a matter problem matched with the target problem in the matter problem set, so that the calculation amount can be reduced, and the searching efficiency and the searching accuracy are improved.

Referring to fig. 5, fig. 5 is a flowchart illustrating another method for intelligent question answering in the embodiment of the present application, including the following steps:

step S500, acquiring a target problem proposed by a target object;

step S501, identifying the item type corresponding to the target problem through a text convolution neural network;

step S502, obtaining a question-answer relationship pair corresponding to the item type in a question-answer knowledge base;

step S503, searching a target answer corresponding to the target question in the question-answer relation pair;

step S504, determining whether a target answer corresponding to the target question is found in the question-answer relationship pair, if yes, performing step S508, otherwise performing step S505;

step S505, performing word segmentation processing on the target problem to obtain each word segmentation in the target problem, and performing feature vector extraction on each word segmentation in the target problem through an ALBert model to obtain corresponding first semantic features;

step S506, based on the first semantic features, retrieving a standard question with the semantic similarity reaching a similarity threshold value with the target question from a preset question-answering library;

step S507, determining whether a standard question is retrieved from a preset question-answering library, if yes, executing step S509, otherwise executing step S508;

step S508, recording the target problem in the message log;

and step S509, feeding back a target answer corresponding to the target question.

According to the method and the device, the target answers are directly returned after the target answers are obtained from the knowledge question-answering library, the searching efficiency is improved, the target answers are not obtained from the knowledge question-answering library, the first semantic features corresponding to the target questions are extracted by combining the whole statement information of the target questions instead of only referring to a single keyword, the identification accuracy is improved, the target answers are directly returned after the target answers are retrieved, and the searching efficiency is improved.

Based on the same inventive concept, an embodiment of the present application provides an intelligent question answering device, please refer to fig. 6, and fig. 6 exemplarily provides an intelligent question answering device 600 in the embodiment of the present application, which includes:

a first obtaining module 601, configured to obtain a target problem proposed by a target object;

the extracting module 602 is configured to, after a target answer corresponding to a target question is not found in a question-answer knowledge base storing a question-answer relationship pair, perform feature vector extraction on each participle in the target question based on context statement information corresponding to each participle in the target question to obtain a corresponding first semantic feature;

the retrieval module 603 is configured to retrieve, based on the first semantic feature, a standard question from a preset question-answer library, where semantic similarity between the standard question and the target question reaches a similarity threshold;

the second obtaining module 604 is configured to obtain a first standard answer corresponding to the standard question, and use the first standard answer as a target answer corresponding to the target question.

In one possible implementation manner, the extracting module 602 searches the question-answer knowledge base for the target answer corresponding to the target question by the following method:

performing word segmentation processing on the target problem, and determining key words and key part-of-speech information in the target problem;

matching the target question with a question template in a question-answer knowledge base based on the keyword and the part-of-speech information, and obtaining a matching result;

and searching a target answer corresponding to the target question in a question-answer knowledge base based on the matching result.

In a possible implementation manner, the extracting module 602 is specifically configured to:

if the matching result represents that the target question template matched with the target question exists, acquiring a second standard answer corresponding to the target question template in a question-answer knowledge base, and taking the second standard answer as a target answer corresponding to the target question; or

And if the matching result represents that the target question template matched with the target question does not exist, determining that the target answer corresponding to the target question is not found in the question and answer knowledge base.

In a possible implementation manner, after obtaining the target question posed by the target object, before searching the question-answer knowledge base of the question-answer relationship pair for the target answer corresponding to the target question, the extracting module 602 is further configured to:

identifying the item type corresponding to the target problem through a text convolution neural network;

searching a question and answer knowledge base for a matter question set corresponding to the matter type;

and searching the item problem matched with the target problem in the item problem set.

In a possible implementation manner, the extracting module 602 is specifically configured to:

extracting feature vectors of all the participles in the target problem through an ALBert model to obtain semantic features corresponding to the target problem;

the retrieving module 603 is specifically configured to:

aiming at any preset problem in a preset question-answer library, determining semantic similarity between the preset problem and a target problem through an Euclidean distance algorithm;

and searching the preset questions with the semantic similarity reaching the similarity threshold value from all the preset questions stored in the preset question-answer library based on the semantic similarity, and taking the preset questions obtained by searching as standard questions.

In a possible implementation manner, the second obtaining module 604 is specifically configured to:

if the number of the obtained standard questions exceeds the number threshold, screening out the standard questions with the semantic similarity arrangement serial numbers larger than the set serial number from the obtained standard questions, and taking all first standard answers corresponding to the standard questions as target answers; or

If the number of the obtained standard questions exceeds the number threshold, extracting a set number of first standard answers from first standard answers corresponding to any standard question, and taking the extracted first standard answers as target answers.

In one possible implementation manner, if a standard question is not retrieved from a preset question-answer library based on a first semantic feature, recording the first semantic feature and the number of questions corresponding to the first semantic feature in a message log;

and when the questioning times reach a time threshold value, issuing the target questions corresponding to the first semantic features to update a preset questioning and answering library.

In some possible embodiments, the aspects of the intelligent question answering method provided in the present application may also be implemented in the form of a program product including program code for causing a computer device to perform the steps of the intelligent question answering method according to various exemplary embodiments of the present application described above in this specification when the program product is run on the computer device.

The program product may employ any combination of each or more of the readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having each or more wires, a portable disk, 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 disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

The program product of the intelligent question answering method of the embodiment of the application can adopt a portable compact disk read only memory (CD-ROM) and comprises program codes, and can be run on the intelligent question answering device.

A readable signal medium may include a data signal propagating in baseband or as a submodel to a carrier wave, in which readable program code is carried. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A readable signal medium may also be any readable medium that is not a readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with a command execution system, apparatus, or device.

Program code embodied on a readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Program code for carrying out operations of the present application may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages.

It should be noted that although in the above detailed description several units or sub-units of the device are mentioned, such division is merely exemplary and not mandatory. Indeed, according to embodiments of the present application, the feature vectors and functions of two or more units described above may be embodied in each unit. Conversely, the feature vectors and functions of each unit described above may be further divided into embodiments by a plurality of units.

Further, while the operations of the methods of the present application are depicted in the drawings in a particular order, this does not require or imply that these operations must be performed in this particular order, or that all of the illustrated operations must be performed, to achieve desirable results. Additionally or alternatively, certain steps may be omitted, multiple steps combined into each step execution, and/or each step broken down into multiple step executions.

It will be apparent to those skilled in the art that various changes and modifications may be made in the present application without departing from the spirit and scope of the application. Thus, if such modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is intended to include such modifications and variations as well.

Claims (10)

1. An intelligent question answering method is characterized by comprising the following steps:

acquiring a target problem proposed by a target object;

after a target answer corresponding to the target question is not found in a question-answer knowledge base storing question-answer relationship pairs, extracting feature vectors of all participles in the target question based on context statement information corresponding to all the participles in the target question to obtain corresponding first semantic features;

searching a standard question with the semantic similarity reaching a similarity threshold value with the target question in a preset question-answer base based on the first semantic features;

and acquiring a first standard answer corresponding to the standard question, and taking the first standard answer as a target answer corresponding to the target question.

2. The method of claim 1, wherein the question-answer knowledge base is searched for the target answer corresponding to the target question by:

performing word segmentation processing on the target question, and determining key words in the target question and part-of-speech information of the key words;

matching the target question with a question template in the question-answer knowledge base based on the keyword and the part-of-speech information, and obtaining a matching result;

and searching a target answer corresponding to the target question in the question-answer knowledge base based on the matching result.

3. The method according to claim 2, wherein said searching for the target answer corresponding to the target question in the question-answer knowledge base based on the matching result specifically comprises:

if the matching result represents that a target question template matched with the target question exists, acquiring a second standard answer corresponding to the target question template in the question-answer knowledge base, and taking the second standard answer as a target answer corresponding to the target question; or

And if the matching result represents that a target question template matched with the target question does not exist, determining that a target answer corresponding to the target question is not found in the question-answer knowledge base.

4. The method according to claim 1, wherein after obtaining the target question posed by the target object, before searching the question-answer knowledge base storing the question-answer relationship pair for the target answer corresponding to the target question, further comprising:

identifying the item type corresponding to the target problem through a text convolution neural network;

searching a question and answer knowledge base for a matter question set corresponding to the matter type;

and searching the item problem matched with the target problem in the item problem set.

5. The method of claim 1, wherein the extracting feature vectors for each participle in the target question to obtain a corresponding first semantic feature comprises:

extracting feature vectors of all the participles in the target problem through an ALBert model to obtain corresponding first semantic features;

the searching, based on the first semantic features, for the standard question with the semantic similarity reaching a similarity threshold with the target question in a preset question-answering library includes:

aiming at any preset question in the preset question-answer library, determining semantic similarity between the preset question and the target question through an Euclidean distance algorithm based on the first semantic feature and a second semantic feature corresponding to the preset question;

and searching the preset questions with the semantic similarity reaching a similarity threshold from all the preset questions stored in the preset question-answer base based on the semantic similarity, and taking the preset questions obtained through searching as the standard questions.

6. The method of claim 1, wherein the obtaining a first standard answer corresponding to the standard question and using the first standard answer as a target answer of the target question comprises:

if the number of the obtained standard problems is determined to exceed the number threshold, screening out the standard problems with the semantic similarity arrangement serial numbers larger than the set serial numbers from the obtained standard problems, and taking all first standard answers corresponding to the screened standard problems as the target answers; or

If the number of the obtained standard questions exceeds the number threshold, extracting a set number of first standard answers from all first standard answers corresponding to any standard question, and taking the extracted first standard answers as the target answers.

7. The method of claim 1, further comprising:

if the standard question is not retrieved from the preset question-answer library based on the first semantic feature, recording the first semantic feature and question times corresponding to the first semantic feature in a message log;

and when the number of times of questioning reaches a threshold value, issuing the target question corresponding to the first semantic feature to update the preset question-answer library.

8. An intelligent question-answering apparatus, comprising: human-computer interaction interface, memory and processor, wherein:

the man-machine interaction interface is used for acquiring a target problem proposed by a target object;

the memory for storing a computer program operable on the processor;

the processor is used for reading the computer program in the memory and executing: after a target answer corresponding to the target question is not found in a question-answer knowledge base storing question-answer relationship pairs, extracting feature vectors of all participles in the target question based on context statement information corresponding to all the participles in the target question to obtain corresponding first semantic features; searching a standard question with the semantic similarity reaching a similarity threshold value with the target question in a preset question-answer base based on the first semantic features; and acquiring a first standard answer corresponding to the standard question, and taking the first standard answer as a target answer corresponding to the target question.

9. An intelligent question answering device, comprising:

the first acquisition module is used for acquiring a target problem proposed by a target object;

the extraction module is used for extracting feature vectors of all the participles in the target question based on context statement information corresponding to all the participles in the target question to obtain corresponding first semantic features after a target answer corresponding to the target question is not found in a question-answer knowledge base storing question-answer relation pairs;

the retrieval module is used for retrieving a standard question with the semantic similarity reaching a similarity threshold value with the target question from a preset question-answering library based on the first semantic features;

and the second acquisition module is used for acquiring a first standard answer corresponding to the standard question and taking the first standard answer as a target answer corresponding to the target question.

10. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program which, when being executed by a processor, carries out the method steps of any one of claims 1 to 7.

Images