CN109726274B - Question generation method, device and storage medium - Google Patents

Abstract

The embodiment of the invention provides a problem generation method, a problem generation device and a computer readable storage medium. The problem generation method comprises the following steps: identifying the text type of the document to be processed according to the text structure; selecting a generative model corresponding to the text type, the generative model comprising at least one of an explicit problem generative model, a structured and semi-structured problem generative model, and a natural language problem generative model; generating a question for the document to be processed using the selected generative model. According to the method and the device, aiming at the characteristics of different text types, for the whole document or all parts of texts of the whole document, the most applicable generation model is selected, and the accuracy rate of generation problems is improved.

Description

Problem generation method, device and storage medium

technical field

The present invention relates to the field of information technology, and in particular, to a problem generation method, device and computer-readable storage medium.

Background technique

FAQ (Frequently Asked Questions, question and answer system) is the main means of providing online help on the current network. By organizing some question-and-answer pairs that may be frequently asked in advance, they are published on web pages to provide users with consulting services.

The FAQ implementation methods of the prior art mainly include the following:

(1) General question answering system, based on retrieval or knowledge-based question answering service.

(2) Customized retrieval, creating an index for document content segmentation and word segmentation; or obtaining question-answer pairs by document structuring or manual screening.

(3) Question retrieval based on word matching or synonym matching.

The deficiencies of the prior art mainly include the following aspects:

(1) Retrieval-based or knowledge-based general question answering systems cannot meet the needs of customization.

(2) For the question-and-answer method by creating an index on the document content, firstly, not all the content is question-and-answer content, so storing the entire text will cause a waste of storage space; secondly, the accuracy of generating questions in this way is low, because words A hit does not mean that the current content is the answer; there is also an inability to determine the answer boundary and an inability to form a visual FAQ document. The so-called visualization refers to the reading comprehension of the depth of the text content, so as to extract several question-and-answer pairs, which is convenient for users to find the question and retrieve the answer. The current technology cannot deeply understand the text or generate a good text.

(3) Question retrieval based on synonym matching or word matching has poor generalization ability and low recall rate.

SUMMARY OF THE INVENTION

Embodiments of the present invention provide a problem generation method, apparatus, and computer-readable storage medium, so as to at least solve one or more technical problems in the prior art.

In a first aspect, an embodiment of the present invention provides a problem generation method, including:

Identify the text type of the document to be processed according to the text structure;

selecting a generative model corresponding to the text type, the generative model includes at least one of an explicit question generation model, a structured and semi-structured question generation model, and a natural language question generation model;

A question is generated for the document to be processed using the selected generative model.

In one embodiment, identifying the text type of the document to be processed according to the text structure includes: identifying whether there is a question-and-answer structure in the text structure of the document to be processed;

Selecting the generation model corresponding to the text type includes: if there is a question-and-answer structure in the text structure of the document to be processed, then using the explicit question generation model as the generation model corresponding to the text type;

Generating a question for the document to be processed using the selected generation model includes: generating a question for the document to be processed by using the explicit question generation model.

In one embodiment, generating a question for the document to be processed using the explicit question generation model includes:

Judging whether the question part in the question-and-answer structure matches the corresponding answer part, and screening out the part of the text corresponding to the question-and-answer structure that has been successfully matched as candidate text;

classifying the screened candidate texts using the first recurrent neural network model to identify explicit questions from the candidate texts;

Treat the explicit question as a question generated for the document to be processed.

In one embodiment, identifying the text type of the document to be processed according to the text structure includes: identifying whether there is a title structure in the text structure of the document to be processed, and the title structure includes a title or a table;

Selecting a generation model corresponding to the text type includes: if there is a title structure in the text structure of the document to be processed, then using the structured and semi-structured problem generation models as the generation model corresponding to the text type ;

Generating a question for the document to be processed using the selected generation model includes: generating a question for the document to be processed using the structured and semi-structured question generation model.

In one embodiment, generating a question for the document to be processed using the structured and semi-structured question generation model includes:

If there is a title in the text structure of the document to be processed, obtain a paraphrase of the attribute related to the title;

A question is generated based on the attribute rephrase.

In one embodiment, obtaining a paraphrase of attributes related to the title includes:

Obtain the search click presentation log related to the title;

Data mining is performed on the search click presentation log to obtain a retelling of attributes related to the title;

The attribute repetition is stored in an attribute repetition table.

In one embodiment, generating a question based on the attribute paraphrase includes:

generating a question using the first encoder-decoder model based on the attribute recap; or,

Query the attribute paraphrase related to the title from the property paraphrase table, and generate a question according to the property paraphrase found.

In one embodiment, identifying the text type of the document to be processed according to the text structure includes: identifying whether there is a question-and-answer structure and a title structure in the text structure of the document to be processed, and the title structure includes a title or a table;

Selecting a generation model corresponding to the text type includes: if there is no question-and-answer structure and no title structure in the text structure of the document to be processed, using the natural language question generation model as a generation model corresponding to the text type ;

Generating a question for the document to be processed using the selected generation model includes: generating a question for the document to be processed by using the natural language question generation model.

In one embodiment, generating a question for the document to be processed using the natural language question generation model includes:

Use the second recurrent neural network model to screen out target sentences from the documents to be processed, where the target sentences include sentences with complete semantics;

using a third recurrent neural network model to select candidate answer segments from the target sentence;

Based on the candidate answer segments, questions are generated using a second encoder-decoder model.

In one embodiment, the method further includes: positioning an answer boundary for the generated question.

In one embodiment, answer boundary positioning for the generated question includes:

Use the bidirectional attention flow network to predict the start and end positions of the answer segments corresponding to the question;

The answer segments are sorted by a learning sorting model, and the answer boundary is located for the question according to the sorting result, wherein the features of the learning sorting model include the start and end positions of the answer segments.

In a second aspect, an embodiment of the present invention provides a problem generation device, including:

A text type identification unit, used to identify the text type of the document to be processed according to the text structure;

a generative model selection unit for selecting a generative model corresponding to the text type, the generative model includes at least one of an explicit question generation model, a structured and semi-structured question generation model, and a natural language question generation model;

A question generating unit, configured to generate a question for the document to be processed by using the selected generation model.

In one embodiment, the text type identification unit includes a first identification subunit, and the first identification subunit is used to: identify whether there is a question-and-answer structure in the text structure of the document to be processed;

The generation model selection unit includes a first selection subunit, and the first selection subunit is used for: if there is a question-and-answer structure in the text structure of the document to be processed, the explicit question generation model is used as the The generative model corresponding to the text type;

The question generation unit includes a first generation subunit, and the first generation subunit is configured to: generate a question for the document to be processed by using the explicit question generation model.

In one embodiment, the first generation subunit is further used for:

Judging whether the question part in the question-and-answer structure matches the corresponding answer part, and screening out the part of the text corresponding to the question-and-answer structure that has been successfully matched as candidate text;

classifying the screened candidate texts using the first recurrent neural network model to identify explicit questions from the candidate texts;

Treat the explicit question as a question generated for the document to be processed.

In one embodiment, the text type identification unit includes a second identification subunit, and the second identification subunit is configured to: identify whether there is a title structure in the text structure of the document to be processed, and the title structure includes headings or tables;

The generation model selection unit includes a second selection subunit, and the second selection subunit is used for: if there is a title structure in the text structure of the document to be processed, generate a model for the structured and semi-structured problems as a generative model corresponding to the text type;

The question generation unit includes a second generation subunit, and the second generation subunit is configured to: generate a question for the document to be processed by using the structured and semi-structured question generation model.

In one embodiment, the second generating subunit includes:

A paraphrase acquisition subunit, used for acquiring a paraphrase of an attribute related to the title when there is a title in the text structure of the document to be processed;

The paraphrase question generation subunit is used to paraphrase the generation question according to the attribute.

In one embodiment, the paraphrase acquisition subunit is further used for:

Obtain the search click presentation log related to the title;

Data mining is performed on the search click presentation log to obtain a retelling of attributes related to the title;

The attribute repetition is stored in an attribute repetition table.

In one embodiment, the paraphrase question generation subunit is further used for:

generating a question using the first encoder-decoder model based on the attribute recap; or,

Query the attribute paraphrase related to the title from the property paraphrase table, and generate a question according to the property paraphrase found.

In one embodiment, the text type identification unit includes a third identification subunit, and the third identification subunit is used for: identifying whether there is a question-and-answer structure and a title structure in the text structure of the document to be processed, the Heading structure including headings or tables;

The generation model selection unit includes a third selection subunit, and the third selection subunit is used for: if there is no question-and-answer structure and no title structure in the text structure of the document to be processed, generate a model for the natural language question as a generative model corresponding to the text type;

The question generating unit includes a third generating subunit, and the third generating subunit is configured to: generate a question for the document to be processed by using the natural language question generating model.

In one embodiment, the third generation subunit is further used for:

Use the second recurrent neural network model to screen out target sentences from the documents to be processed, where the target sentences include sentences with complete semantics;

using a third recurrent neural network model to select candidate answer segments from the target sentence;

Based on the candidate answer segments, questions are generated using a second encoder-decoder model.

In one embodiment, the apparatus further includes an answer boundary locating unit for locating the answer boundary for the generated question.

In one embodiment, the answer boundary positioning unit is further used for:

Use the bidirectional attention flow network to predict the start and end positions of the answer segments corresponding to the question;

The answer segments are sorted by a learning sorting model, and the answer boundary is located for the question according to the sorting result, wherein the features of the learning sorting model include the start and end positions of the answer segments.

In a third aspect, an embodiment of the present invention provides an apparatus for generating a problem. The function of the apparatus may be implemented by hardware, or by executing corresponding software in hardware. The hardware or software includes one or more modules corresponding to the above functions.

In a possible design, the structure of the apparatus includes a processor and a memory, the memory is used to store a program that supports the apparatus to perform the above method, and the processor is configured to execute the memory stored in the memory program of. The apparatus may also include a communication interface for communicating with other devices or a communication network.

In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, which stores a computer program, and when the program is executed by a processor, implements any one of the methods described in the first aspect above.

One of the above technical solutions has the following advantages or beneficial effects: according to the characteristics of different text types, for the entire document, or for each part of the text of the entire document, the most suitable generation model is selected, Improves the accuracy of generating questions.

Another technical solution in the above technical solutions has the following advantages or beneficial effects: the accurate boundary of the answer corresponding to the question can be obtained through the question answering technology, which further improves the accuracy of the generated FAQ document.

The above summary is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, further aspects, embodiments and features of the present invention will become apparent by reference to the accompanying drawings and the following detailed description.

Description of drawings

In the drawings, unless stated otherwise, the same reference numbers refer to the same or like parts or elements throughout the several figures. The drawings are not necessarily to scale. It should be understood that these drawings depict only some embodiments according to the disclosure and should not be considered as limiting the scope of the invention.

FIG. 1 is a flowchart of a problem generation method provided by an embodiment of the present invention.

FIG. 2 is a schematic diagram of a sample FAQ mining target document of the method for generating questions according to an embodiment of the present invention.

FIG. 3 is a schematic diagram of a text type of a question generation method provided by an embodiment of the present invention.

FIG. 4 is a flowchart of generating a question by using an explicit question generating model of the question generating method provided by the embodiment of the present invention.

FIG. 5 is a flowchart of generating a question by using an explicit question generating model of the question generating method provided by an embodiment of the present invention.

FIG. 6 is a flowchart of generating a question using a structured and semi-structured question generating model of the question generating method provided by an embodiment of the present invention.

FIG. 7 is a flowchart of generating a question using a structured and semi-structured question generating model of the question generating method provided by an embodiment of the present invention.

FIG. 8 is a flowchart of generating a question using a structured and semi-structured question generating model of the question generating method provided by an embodiment of the present invention.

FIG. 9 is a flowchart of generating a question using a natural language question generating model of the question generating method provided by an embodiment of the present invention.

FIG. 10 is a flowchart of generating a question using a natural language question generating model of the question generating method provided by an embodiment of the present invention.

FIG. 11 is a flowchart of a problem generation method provided by an embodiment of the present invention.

FIG. 12 is a schematic diagram of the online part of the question generation method provided by the embodiment of the present invention.

FIG. 13 is a flowchart of answer boundary positioning of the question generation method provided by the embodiment of the present invention.

FIG. 14 is a schematic diagram of an offline part of a problem generation method provided by an embodiment of the present invention.

FIG. 15 is a structural block diagram of an apparatus for generating a question provided by an embodiment of the present invention.

FIG. 16 is a structural block diagram of an apparatus for generating a question provided by an embodiment of the present invention.

FIG. 17 is a structural block diagram of a second generating subunit of the question generating apparatus provided by the embodiment of the present invention.

FIG. 18 is a structural block diagram of an apparatus for generating a question provided by an embodiment of the present invention.

FIG. 19 is a structural block diagram of an apparatus for generating a question provided by an embodiment of the present invention.

Detailed ways

In the following, only certain exemplary embodiments are briefly described. As those skilled in the art would realize, the described embodiments may be modified in various different ways, all without departing from the spirit or scope of the present invention. Accordingly, the drawings and description are to be regarded as illustrative in nature and not restrictive.

FIG. 1 is a flowchart of a problem generation method provided by an embodiment of the present invention. As shown in FIG. 1, the problem generation method of the embodiment of the present invention includes:

Step S110, identifying the text type of the document to be processed according to the text structure;

Step S120, selecting a generative model corresponding to the text type, where the generative model includes at least one of an explicit question generation model, a structured and semi-structured question generation model, and a natural language question generation model;

Step S130, using the selected generation model to generate a question for the document to be processed.

FAQs are used on many websites to provide consulting services to users. Some frequently asked questions by users are listed in the FAQ, which is a form of online help. When users use the functions or services of some websites, they often encounter some seemingly simple problems that may be difficult to understand without explanation. Sometimes users are even lost because of these details. In many cases, these problems can be solved with a simple explanation, which is the value of the FAQ.

The questions and answers in the FAQ design must be frequently asked and encountered by users. For example, in network marketing, FAQ is considered as a common online customer service means, a good FAQ system should be able to answer at least 80% of the general and frequently asked questions of users. Through the use of FAQ, it is not only convenient for users, but also greatly reduces the pressure of website staff, saves a lot of customer service costs, and increases user satisfaction. Therefore, in FAQ design, it is crucial to use scientifically sound methods to generate high-accuracy questions.

FIG. 2 is a schematic diagram of a sample FAQ mining target document of the method for generating questions according to an embodiment of the present invention. Question-and-answer pairs can be automatically generated from the specified documents shown in Figure 2. An example of the expected question-and-answer pairs is as follows:

Q: What are the effects of secondhand smoke on the body?

A: 1. Increase the chance of lung cancer...; 2. Harm to memory, nicotine in smoke... 3. Cause childhood asthma, pneumonia, ear inflammation...

Q: How many times is the risk of lung cancer in people who smoke secondhand smoke than normal people?

A: 2.6 times to 6 times.

In the above, Q(question) represents the question, and A(answer) represents the answer corresponding to the question.

The embodiment of the present invention realizes intelligent question and answer through customized retrieval, and can process specified documents to achieve the purpose of customized retrieval, so that the generated intelligent question and answer is more pertinent. Through reading comprehension, question generation technology and question answering technology, the document to be processed is analyzed, the questions that users may ask are extracted, and the customized question answering service is realized.

Specifically, the text of the specified document to be processed can be classified into three text types: explicit FAQ text type, structured and semi-structured text type, and natural language text type. FIG. 3 is a schematic diagram of a table of text types of a question generation method provided by an embodiment of the present invention. "Q" in Figure 3 represents the generated question corresponding to each text sample.

The first row of the table in Figure 3 shows the explicit FAQ text type, and the corresponding generated question is: "Can I order local and domestic traffic packages for unlimited 98 yuan?"

The second row of the table in Figure 3 shows structured and semi-structured text types, and the corresponding generated question is: "What are the ways to apply for Baidu Goddess Card?" In the case of a tabular structure, "semi-structured" may include the case where the document has headings. The "Baidu Goddess Card" in the generated question can be obtained according to the general title of the document or the terms and keywords related to the document.

The third row in the table in Figure 3 shows the natural language text type, and the corresponding generated question is: "How to charge for new 4G Fuka users to access the network?" Among them, "4G Fuka" in the generated question can be Obtained according to the general title of the document or the terms and keywords related to the document.

The embodiment of the present invention selects different generation models corresponding to the text types for documents of different text types, so that the generated questions are more targeted, have higher accuracy, and are more in line with the needs of customized question answering services. The generated questions and their answers can be used to support customer service robots to achieve human-computer interaction.

In the above technical solution, an index can also be created for the questions to save storage space. For example, the manner of creating an index may include an inverted index or a key value index. Among them, the inverted index originates from the situation in which the records need to be searched according to the value of the attribute in practical applications. Each entry in such an index table includes an attribute value and the address of each record with that attribute value. Because the attribute value is not determined by the record, but the position of the record is determined by the attribute value, so it is called an inverted index. The recall accuracy of the query answers based on the question index is higher. Further, the generated question and answer pairs can also be used to establish a semantic index to support the customer service robot to achieve human-computer interaction.

In an example, the document to be processed may be one of the above three text types for the entire document, and a generation model corresponding to the text type may be selected to generate the problem for the entire document. In another example, the document to be processed may be divided into several parts, and the text types of the texts of each part belong to one of the above three text types respectively. In this case, the generation model corresponding to the text type of each part of the text can be selected respectively, and the question can be generated for each part of the text.

In one embodiment, identifying the text type of the document to be processed according to the text structure includes: identifying the text type of each part of the text in the document to be processed according to the text structure;

Selecting a generation model corresponding to the text type includes: selecting a generation model corresponding to the text type of each part of the text;

Using the selected generative model to generate a question for the document to be processed includes: generating a question for each part of the text by using the selected generative model.

First, identify which text type the document to be processed or each part of the document to process belongs to according to the text structure:

(1) Identify whether there is a question-and-answer structure in the text structure of the document to be processed, that is, questions and answers to questions. If there is, an explicit question generation model is used to generate questions for the part of the text with a question-and-answer structure in the document to be processed.

(2) Identify whether there is a title or table in the text structure of the document to be processed, and if so, use a structured and semi-structured problem generation model to generate questions for part of the text with a title or table in the document to be processed.

(3) If part of the text in the document to be processed has no question-and-answer structure, no title or table, for example, part of the text in the document to be processed is in the format of a plain text document, the part of the text can be called a plain text part. In this case, a natural language question generation model is used to generate questions for this part of the text.

The above technical solution has the following advantages or beneficial effects: according to the characteristics of different text types, for the entire document or for each part of the entire document, the most suitable generation model is selected, which improves the generation problem. Accuracy.

FIG. 4 is a flowchart of generating a question by using an explicit question generating model of the question generating method provided by the embodiment of the present invention. As shown in FIG. 4 , in one embodiment, in step S110 in FIG. 1 , identifying the text type of the document to be processed according to the text structure may specifically include step S210 : identifying whether there is a question and answer in the text structure of the document to be processed structure.

Step S120 in FIG. 1 selects a generation model corresponding to the text type, which may specifically include step S220: if there is a question-and-answer structure in the text structure of the document to be processed, the explicit question generation model is used as the The generative model corresponding to the text type described.

Step S130 in FIG. 1 , using the selected generation model to generate a question for the document to be processed, may specifically include step S230 : using the explicit question generation model to generate a question for the document to be processed.

Referring to the example shown in the first row of the table in FIG. 3 , if there is a question and answer structure in the text structure of the document to be processed, the text type of this part of the text belongs to the explicit FAQ text type. For the explicit FAQ text type, select the corresponding explicit question generation model to generate questions for this part of the text.

FIG. 5 is a flowchart of generating a question by using an explicit question generating model of the question generating method provided by an embodiment of the present invention. As shown in FIG. 5 , in one embodiment, in step S230 in FIG. 4 , the explicit question generation model is used to generate questions for the document to be processed, which may specifically include:

Step S310, judging whether the question part in the question-and-answer structure matches the corresponding answer part, and selecting the part of the text corresponding to the question-and-answer structure that has been successfully matched as candidate text;

Step S320, using the first recurrent neural network model to classify the screened candidate texts to identify explicit questions from the candidate texts;

Step S330, taking the explicit question as a question generated for the document to be processed.

In such an embodiment, questions may be generated through document structure parsing and question identification. Among them, the process of document structure parsing may include preliminarily screening out possible explicit problems with the text structure. The process of problem identification may include: using RNN (Recurrent Neural Network, Recurrent Neural Network) as a problem classification model, using words as features, using artificially labeled training data, and training the above RNN model as an explicit problem generation model.

Specifically, the processing process of the explicit problem generation model can be divided into the following steps:

(1) Document structure analysis: find out the question part and the corresponding answer part in the text, and judge whether the question part and the answer part match. If the question part and the answer part match, the question part and the answer part are filtered out as "Possible Explicit Questions".

For example, the method of "finding the question part and the corresponding answer part in the text" may include: finding the question mark, determining the sentence before the question mark as the question part, and determining the paragraph after the question mark as the answer part. Semantic understanding can be used to determine whether the question and answer match.

(2) Problem identification: Use the RNN model, that is, the above-mentioned first recurrent neural network model, to classify the screened "possible explicit problems". The results output by the first recurrent neural network model can be divided into two categories, one is to finally determine that it is an explicit problem, and the other is to finally determine that it is not an explicit problem.

The features used by the above RNN model may include: part-of-speech and word dependencies. Among them, the text can be divided into sentences and words, and different features can be extracted as the features of the training model according to the needs. For example, use punctuation for sentence segmentation; use NLP (Natural Language Processing, natural language processing) for word segmentation. You can try to divide a sentence into patterns such as 5 words, 3 words or 1 word respectively, and select the pattern with the best effect for word segmentation.

FIG. 6 is a flowchart of generating a question using a structured and semi-structured question generating model of the question generating method provided by an embodiment of the present invention. As shown in FIG. 6 , in one embodiment, in step S110 in FIG. 1 , identifying the text type of the document to be processed according to the text structure may specifically include step S410 : identifying whether there is a title in the text structure of the document to be processed structure, which includes headings or tables.

Step S120 in FIG. 1 selects a generation model corresponding to the text type, which may specifically include step S420: if the text structure of the document to be processed has a title structure, generate the structured and semi-structured questions The model acts as a generative model corresponding to the text type.

Step S130 in FIG. 1 , using the selected generation model to generate a question for the document to be processed, may specifically include step S430 : using the structured and semi-structured question generation model to generate a question for the document to be processed.

Referring to the example shown in the second row of the table in FIG. 3 , if the text structure of the document to be processed has a title structure, where the title structure includes a title or a table, the text type of this part of the text belongs to structured and semi-structured text types. For structured and semi-structured text types, select the corresponding structured and semi-structured question generation models to generate questions for this part of the text.

FIG. 7 is a flowchart of generating a question using a structured and semi-structured question generating model of the question generating method provided by an embodiment of the present invention. As shown in FIG. 7 , in one embodiment, in step S430 in FIG. 6 , the structured and semi-structured question generation models are used to generate questions for the documents to be processed, which may specifically include:

Step S510, in the case that there is a title in the text structure of the document to be processed, obtain an attribute paraphrase related to the title;

Step S520, generating a question according to the attribute paraphrase.

In this embodiment, the search click presentation log in the FAQ retrieval system can be obtained, and the retelling of the properties of the search click presentation log is mined by means of the search. For example: *Billing method ->*How to charge, these two attributes can be repeated for each other. Generation of attribute-based paraphrases as semi-structured and structured question generative models.

FIG. 8 is a flowchart of generating a question using a structured and semi-structured question generating model of the question generating method provided by an embodiment of the present invention. As shown in FIG. 8 , in one embodiment, in step S510 in FIG. 7 , acquiring the attribute paraphrase related to the title may specifically include:

Step S610, obtaining a search click presentation log related to the title;

Step S620, performing data mining on the search click presentation log to obtain a retelling of attributes related to the title;

Step S630, storing the attribute paraphrase in the attribute paraphrase table.

In one embodiment, step S520 in FIG. 7 , generating a question according to the attribute retelling, may specifically include:

generating a question using the first encoder-decoder model based on the attribute recap; or,

Query the attribute paraphrase related to the title from the property paraphrase table, and generate a question according to the property paraphrase found.

Specifically, structured and semi-structured generative models may include the following processing steps:

(1) Rely on the click display log of the search, and use the method of data mining to mine the repetition of the attribute. Stores the rephrases of the mined attributes in the attribute rehearsal table.

For example, two users A and B click on the same URL (Uniform Resource Locator) after using different keywords to search. Then the meanings expressed by these two different keywords may be the same and can be repeated for each other.

In one example, a user searches for "efficacy of bitter gourd." Among them, "bitter gourd" is the entity, and "efficacy" is the attribute of the entity. The repetition of the attribute "efficacy" also includes "action", "pharmaceutical effect" and so on. That is to say, "the effect of bitter gourd", "the effect of bitter gourd" and "the medicinal effect of bitter gourd" have the same meaning.

(2) Attribute-based paraphrase generation as a semi-structured and structured question generation model. Two implementations can be included:

Mode 1: Use the Seq2Seq (Sequence to Sequence, sequence to sequence) model, that is, the above-mentioned first encoder-decoder model, to generate the problem. The features used by the above models include lexical and syntactic features, the start and end positions of the answers predicted by the sequence tagging model, and word features. The input information of the above model is the paragraph of the document to be processed, and the output information is the problem generated for the input information. In one example, the text content in the document to be processed is: "Beijing is the capital of China." Then the sequence labeling model can label "Beijing". Then, "Beijing" and "Beijing is the capital of China" are used as input information to the seq2seq model to generate the question "where is the capital of China".

The Seq2Seq model, also known as the Encoder-Decoder model (encoder-decoder model), is an important variant of the RNN model. The Encoder-Decoder structure does not limit the sequence length of input and output, so it has a wide range of applications, such as machine translation, text summarization, reading comprehension, speech recognition, etc.

Since the seq2seq model can have unequal lengths of input and output sequences, that is, Sequence to Sequence, it realizes the conversion from one sequence to another, for example, it can implement a chatbot dialogue model. While the classic RNN model fixes the size of the input and output sequences, the seq2seq model breaks this limitation.

The encoder and decoder correspond to the two RNNs of the input sequence and the output sequence, respectively. The basic idea of the common encoder-decoder structure is to use two RNNs, one RNN as the encoder and the other RNN as the decoder. The encoder is responsible for compressing the input sequence into a vector of a specified length. This vector can be regarded as the semantics of the sequence. This process is called encoding. The decoder is responsible for generating the specified sequence according to the semantic vector, which is also called decoding.

Method 2: Query the attribute paraphrase table and select a relevant paraphrase to generate the question.

For example, for the queried paraphrases: "the effect of bitter gourd", "the effect of bitter gourd", and "the medicinal effect of bitter gourd", one of the most suitable paraphrases can be selected from among them to generate the question. By means of semantic understanding, keyword matching and other methods, each queried paraphrase can be compared and analyzed with the expressions in the document to be processed, so as to determine the most suitable paraphrase for generating the question.

In another case, in the case where there is a table in the text structure of the document to be processed, methods such as semantic understanding and keyword matching can be used to identify the content of the header, each row record, and each column field in the table, and then determine the generated content. questions and their corresponding answers. For example, the content of one column of data in the table may be the generated question, while the content of another column of data may be its corresponding answer.

FIG. 9 is a flowchart of generating a question using a natural language question generating model of the question generating method provided by an embodiment of the present invention. As shown in FIG. 9 , in one embodiment, in step S110 in FIG. 1 , identifying the text type of the document to be processed according to the text structure may specifically include step S710 : identifying whether there is a question and answer in the text structure of the document to be processed structure and heading structure, including headings or tables;

Step S120 in FIG. 1 selects the generation model corresponding to the text type, which may specifically include step S720: if there is no question-and-answer structure and no title structure in the text structure of the document to be processed, generate the natural language question model as a generative model corresponding to the text type;

Step S130 in FIG. 1 , using the selected generation model to generate a question for the document to be processed, may specifically include step S730 : using the natural language question generation model to generate a question for the document to be processed.

Referring to the example shown in the third row of the table in FIG. 3 , if there is no question-and-answer structure and no title structure in the text structure of the document to be processed, the text type of this part of the text belongs to the natural language text type. For the natural language text type, select the corresponding natural language question generation model to generate questions for this part of the text.

FIG. 10 is a flowchart of generating a question using a natural language question generating model of the question generating method provided by the embodiment of the invention. As shown in FIG. 10 , in one embodiment, in step S730 in FIG. 9 , using the natural language question generation model to generate a question for the document to be processed may specifically include:

Step S810, using the second recurrent neural network model to screen out a target sentence from the document to be processed, and the target sentence includes a sentence with complete semantics;

Step S820, using the third recurrent neural network model to select candidate answer segments from the target sentence;

Step S830, using the second encoder-decoder model to generate a question according to the candidate answer segment.

In this embodiment, the RNN model may be used to classify and screen target sentences, and the RNN sequence annotation may be used to select candidate answer segments for the screened target sentences, and then the seq2seq model may be used to generate questions.

Specifically, the natural language generation model may include the following processing steps:

(1) First, use the second recurrent neural network model to classify and filter the target sentence, and screen out the sentences with complete semantics.

(2) For the screened target sentences, the third recurrent neural network model is used to select candidate answer segments, that is, the question and the corresponding segment that may be the answer are selected. Among them, the third recurrent neural network model can be trained by using sequence labeling. Sequence tagging can include sentence tagging, that is, tagging questions that can be asked.

(3) Reuse the seq2seq model, that is, the above-mentioned second encoder-decoder model, and the problem generation model generates the problem.

In one example, the text content in the document to be processed is: "Beijing is the capital of China." Then the sequence annotation can label "Beijing". Then, "Beijing" and "Beijing is the capital of China" are used as input information to the seq2seq model to generate the question "where is the capital of China".

FIG. 11 is a flowchart of a problem generation method provided by an embodiment of the present invention. As shown in FIG. 11 , in an embodiment, the method further includes step S140 : positioning the answer boundary for the generated question.

The above technical solution has the following advantages or beneficial effects: through question answering technology, such as answer boundary positioning, the accurate boundary of the answer corresponding to the question can be obtained, which further improves the accuracy of the generated FAQ document.

In one example, the problem generation method of the embodiment of the present invention includes two parts: the online part of the problem generation method and the offline part of the problem generation method. FIG. 12 is a schematic diagram of the online part of the question generation method provided by the embodiment of the present invention. The "target document" in Fig. 12 is the document to be processed. "General document parsing" in Figure 12 involves identifying document structure. Specifically, "general document parsing" may include identifying the subject (title), subtitle, paragraphs and text under the subtitle of the document, and the identified document structure may be described using a tree structure. "General document parsing" also includes identifying which of the three text types described above (explicit FAQ text type, structured and semi-structured text type, natural language text type) each part of the document to be processed belongs to. After identifying the text type of each part in the document to be processed, in the next step, the corresponding question generation model is selected to generate questions and locate the answer boundary.

FIG. 13 is a flowchart of answer boundary positioning of the question generation method provided by the embodiment of the present invention. As shown in FIG. 13 , in an implementation manner, in step S140 in FIG. 11 , positioning the answer boundary for the generated question may specifically include:

Step S910, using a bidirectional attention flow network to predict the start and end positions of the answer segments corresponding to the question;

Step S920, using a learning sorting model to sort the answer segments, and positioning the answer boundary for the question according to the sorting result, wherein the features of the learning sorting model include the start and end positions of the answer segments.

Specifically, answer boundary localization may include the following processing steps:

(1) Using Bi-DAF (Bi-Directional Attention Flow network, bidirectional attention flow network) as a reading comprehension model can accurately predict the starting and ending positions of the answer.

Bidirectional attention flow network is a hierarchical multi-stage structure that can model context at different levels of granularity. The bidirectional attention flow network mainly consists of modeling the context at the character granularity level (Character Level) and the word granularity level (WordLevel), and using the bidirectional attention flow to obtain the representation of the question-aware context.

An exemplary bidirectional attention flow network may include the following layers:

1. Character embedding layer

The main role of this layer is to map words to a fixed-size vector, and this layer can use a character level convolutional neural network (Character level CNN) to achieve its function.

2. Word embedding layer

A pre-trained word embedding model can be used to map each word to a fixed-size vector.

3. Contextual embedding layer

The main function of this layer is to add a context cue (cue) to each word, and the first three layers are applied to the question and context.

4. Attention flow layer

Combining the question and context vectors produces a question-aware feature vector set.

5. Modeling layer

The context can be scanned using a recurrent neural network.

6. Output layer

This layer provides answers to questions.

(2) LTR (Learning to rank, learning to rank) is used to sort the answer fragments.

The start and end positions predicted in step (1) are used as a feature of the LTR in step (2). Step (2) LTR is used to find the answer corresponding to the question from the large text behind the question. Among them, the LTR model ranks the answer fragments according to the question and answer features.

Among them, learning to rank is a supervised learning ranking method. Using LTR, the correlation function can be constructed to sort according to the correlation. For traditional ranking methods, it is difficult to fuse multiple information, and overfitting is likely to occur. Learning sorting is easy to integrate a variety of features, and has a mature and profound theoretical foundation. The parameters are optimized through iteration, and there is a set of mature theories to solve problems such as sparsity and overfitting.

In this embodiment, the target document is first segmented, and natural paragraphs can be identified or segmented using a list extraction method. The paragraphs are then extracted and ranked, for example, using relevant tools such as "Domain Features" and/or "Match Features". Features may include the following:

Question-answer matching features: alignment matching technology, DNN (Deep Neural Networks, deep neural network) QP matching technology, QP matching technology combined with knowledge graph, where Q represents question (Query), P represents segmented paragraph (Paragrap), alignment Matching includes Q and P alignment;

Domain features: entity question answering feature, how why question answering feature, yes/no question answering feature, descriptive question answering feature;

Structural features: list structural features;

Text features: content quality features;

Cross-validation features: Text aggregation features.

FIG. 14 is a schematic diagram of an offline part of a problem generation method provided by an embodiment of the present invention. As shown in Figure 14, the main generation models and data of the offline part can include two parts and five models. Among them, two parts include document title data and question and answer annotation data. The five models include explicit question generation model, structured and semi-structured question generation model, natural language question generation model, Bi-DAF model (reading comprehension model) and LTR model (answer fragment ranking model).

FIG. 15 is a structural block diagram of an apparatus for generating a question provided by an embodiment of the present invention. As shown in FIG. 15 , the problem generating apparatus according to the embodiment of the present invention includes:

A text type identification unit 100, configured to identify the text type of the document to be processed according to the text structure;

A generative model selection unit 200 for selecting a generative model corresponding to the text type, the generative model includes at least one of an explicit question generation model, a structured and semi-structured question generation model, and a natural language question generation model ;

The question generating unit 300 is configured to generate a question for the document to be processed by using the selected generation model.

FIG. 16 is a structural block diagram of an apparatus for generating a question provided by an embodiment of the present invention. As shown in FIG. 16 , in an implementation manner, the text type identification unit 100 includes a first identification subunit 110, and the first identification subunit 110 is configured to: identify whether the text structure of the document to be processed is Has a question-and-answer structure;

The generation model selection unit 200 includes a first selection subunit 210, and the first selection subunit 210 is configured to: if there is a question-and-answer structure in the text structure of the document to be processed, the explicit question generation model is used as the a generative model corresponding to the text type;

The question generating unit 300 includes a first generating subunit 310, and the first generating subunit 310 is configured to: generate a question for the document to be processed by using the explicit question generating model.

In one embodiment, the first generating subunit 310 is further configured to:

Judging whether the question part in the question-and-answer structure matches the corresponding answer part, and screening out the part of the text corresponding to the question-and-answer structure that has been successfully matched as candidate text;

classifying the screened candidate texts using the first recurrent neural network model to identify explicit questions from the candidate texts;

Treat the explicit question as a question generated for the document to be processed.

In an implementation manner, the text type identification unit 100 includes a second identification subunit 120, and the second identification subunit 120 is configured to: identify whether there is a title structure in the text structure of the document to be processed, the Heading structure including headings or tables;

The generation model selection unit 200 includes a second selection subunit 220, and the second selection subunit 220 is used for: if the text structure of the document to be processed has a title structure, then the structured and semi-structured The question generation model is used as the generation model corresponding to the text type;

The question generating unit 300 includes a second generating subunit 320, and the second generating subunit 320 is configured to: generate a question for the document to be processed by using the structured and semi-structured question generating model.

FIG. 17 is a structural block diagram of a second generating subunit of the question generating apparatus provided by the embodiment of the present invention. As shown in FIG. 17 , in one embodiment, the second generating subunit 320 includes:

A paraphrase acquisition subunit 321, configured to acquire a paraphrase of an attribute related to the title when there is a title in the text structure of the document to be processed;

The paraphrase question generating subunit 322 is configured to paraphrase and generate a question according to the attribute.

In one embodiment, the paraphrase obtaining subunit 321 is further configured to:

Obtain the search click presentation log related to the title;

Data mining is performed on the search click presentation log to obtain a retelling of attributes related to the title;

The attribute repetition is stored in an attribute repetition table.

In one embodiment, the paraphrase question generation subunit 322 is further configured to:

generating a question using the first encoder-decoder model based on the attribute recap; or,

Query the attribute paraphrase related to the title from the property paraphrase table, and generate a question according to the property paraphrase found.

Referring to FIG. 16, in one embodiment, the text type identification unit 100 includes a third identification subunit 130, and the third identification subunit 130 is used to: identify whether there is a question and answer in the text structure of the document to be processed structure and heading structure, including headings or tables;

The generation model selection unit 200 includes a third selection sub-unit 230, and the third selection sub-unit 230 is used for: if there is no question-and-answer structure and no title structure in the text structure of the document to be processed, then the natural language The question generation model is used as the generation model corresponding to the text type;

The question generating unit 300 includes a third generating subunit 330, and the third generating subunit 330 is configured to: generate a question for the document to be processed by using the natural language question generating model.

In one embodiment, the third generating subunit 330 is further configured to:

Use the second recurrent neural network model to screen out target sentences from the documents to be processed, where the target sentences include sentences with complete semantics;

using a third recurrent neural network model to select candidate answer segments from the target sentence;

Based on the candidate answer segments, questions are generated using a second encoder-decoder model.

FIG. 18 is a structural block diagram of an apparatus for generating a question provided by an embodiment of the present invention. As shown in FIG. 18 , in an embodiment, the apparatus further includes an answer boundary locating unit 400 for locating the answer boundary for the generated question.

In one embodiment, the answer boundary locating unit 400 is further configured to:

Use the bidirectional attention flow network to predict the start and end positions of the answer segments corresponding to the question;

The answer segments are sorted by a learning sorting model, and the answer boundary is located for the question according to the sorting result, wherein the features of the learning sorting model include the start and end positions of the answer segments.

For the functions of each unit in the problem generating apparatus according to the embodiment of the present invention, reference may be made to the relevant description of the above method, and details are not described herein again.

In a possible design, the structure of the problem generating apparatus includes a processor and a memory, the memory is used for storing a program that supports the problem generating apparatus to execute the above-mentioned problem generating method, and the processor is configured to execute the memory program stored in. The question generating apparatus may further include a communication interface, and the question generating apparatus communicates with other devices or a communication network.

FIG. 19 is a structural block diagram of an apparatus for generating a question provided by an embodiment of the present invention. As shown in FIG. 19 , the apparatus includes: a memory 101 and a processor 102 , and a computer program that can be executed on the processor 102 is stored in the memory 101 . When the processor 102 executes the computer program, the problem generation method in the above-mentioned embodiment is implemented. The number of the memory 101 and the processor 102 may be one or more.

The device also includes:

The communication interface 103 is used to communicate with external devices and perform data interactive transmission.

The memory 101 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk memory.

If the memory 101, the processor 102, and the communication interface 103 are independently implemented, the memory 101, the processor 102, and the communication interface 103 can be connected to each other through a bus and complete communication with each other. The bus may be an Industry Standard Architecture (ISA, Industry Standard Architecture) bus, a Peripheral Component Interconnect (PCI, Peripheral Component) bus or an Extended Industry Standard Architecture (EISA, Extended Industry Standard Component) bus or the like. The bus can be divided into an address bus, a data bus, a control bus, and the like. For ease of presentation, only one thick line is shown in FIG. 19, but it does not mean that there is only one bus or one type of bus.

Optionally, in terms of specific implementation, if the memory 101, the processor 102 and the communication interface 103 are integrated on one chip, the memory 101, the processor 102 and the communication interface 103 can communicate with each other through an internal interface.

In another aspect, an embodiment of the present invention provides a computer-readable storage medium, which stores a computer program, and when the program is executed by a processor, implements any one of the above-mentioned problem generation methods.

In the description of this specification, description with reference to the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples", etc., mean specific features described in connection with the embodiment or example , structure, material or feature is included in at least one embodiment or example of the present invention. Furthermore, the particular features, structures, materials or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, those skilled in the art may combine and combine the different embodiments or examples described in this specification, as well as the features of the different embodiments or examples, without conflicting each other.

In addition, the terms "first" and "second" are only used for descriptive purposes, and should not be construed as indicating or implying relative importance or implying the number of indicated technical features. Thus, a feature delimited with "first", "second" may expressly or implicitly include at least one of that feature. In the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

Any description of a process or method in the flowcharts or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing a specified logical function or step of the process , and the scope of the preferred embodiments of the invention includes alternative implementations in which the functions may be performed out of the order shown or discussed, including performing the functions substantially concurrently or in the reverse order depending upon the functions involved, which should It is understood by those skilled in the art to which the embodiments of the present invention belong.

The logic and/or steps represented in flowcharts or otherwise described herein, for example, may be considered an ordered listing of executable instructions for implementing the logical functions, may be embodied in any computer-readable medium, For use with, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other system that can fetch instructions from and execute instructions from an instruction execution system, apparatus, or apparatus) or equipment. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport the program for use by or in connection with an instruction execution system, apparatus, or apparatus. More specific examples (non-exhaustive list) of computer readable media include the following: electrical connections with one or more wiring (electronic devices), portable computer disk cartridges (magnetic devices), random access memory (RAM), Read Only Memory (ROM), Erasable Editable Read Only Memory (EPROM or Flash Memory), Fiber Optic Devices, and Portable Read Only Memory (CDROM). In addition, the computer readable medium may even be paper or other suitable medium on which the program may be printed, as the paper or other medium may be optically scanned, for example, followed by editing, interpretation, or other suitable medium as necessary process to obtain the program electronically and then store it in computer memory.

It should be understood that various parts of the present invention may be implemented in hardware, software, firmware or a combination thereof. In the above-described embodiments, various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented by any one or a combination of the following techniques known in the art: Discrete logic circuits, application specific integrated circuits with suitable combinational logic gates, Programmable Gate Arrays (PGA), Field Programmable Gate Arrays (FPGA), etc.

Those skilled in the art can understand that all or part of the steps carried by the methods of the above embodiments can be completed by instructing the relevant hardware through a program, and the program can be stored in a computer-readable storage medium, and the program can be stored in a computer-readable storage medium. When executed, one or a combination of the steps of the method embodiment is included.

In addition, each functional unit in each embodiment of the present invention may be integrated into one processing module, or each unit may exist physically alone, or two or more units may be integrated into one module. The above-mentioned integrated modules can be implemented in the form of hardware, and can also be implemented in the form of software function modules. If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they may also be stored in a computer-readable storage medium. The storage medium may be a read-only memory, a magnetic disk or an optical disk, and the like.

The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited to this. Any person skilled in the art who is familiar with the technical field disclosed in the present invention can easily think of various changes or Replacement, these should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims (24)

1. a problem generation method, is characterized in that, comprises: Identify the text type of the document to be processed according to the text structure, the text type includes at least one of an explicit FAQ text type, a structured and semi-structured text type, and a natural language text type; Select a generative model corresponding to the text type, the generative model includes at least one of an explicit question generation model, a structured and semi-structured question generation model, and a natural language question generation model; when the text type is explicit In the case of the type of FAQ text, the explicit question generation model is used as the generation model corresponding to the text type; when the text type is structured and semi-structured text type, the structured and the semi-structured question generation model as the generation model corresponding to the text type; in the case that the text type is a natural language text type, the natural language question generation model is used as the generation model corresponding to the text type ; A question is generated for the document to be processed using the selected generative model. 2. The method according to claim 1, wherein Identifying the text type of the document to be processed according to the text structure includes: identifying whether there is a question-and-answer structure in the text structure of the document to be processed; Selecting the generation model corresponding to the text type includes: if there is a question-and-answer structure in the text structure of the document to be processed, then using the explicit question generation model as the generation model corresponding to the text type; Generating a question for the document to be processed using the selected generation model includes: generating a question for the document to be processed by using the explicit question generation model. 3. The method according to claim 2, wherein generating a question for the document to be processed using the explicit question generation model, comprising: Judging whether the question part in the question-and-answer structure matches the corresponding answer part, and screening out the part of the text corresponding to the question-and-answer structure that has been successfully matched as candidate text; classifying the screened candidate texts using the first recurrent neural network model to identify explicit questions from the candidate texts; Treat the explicit question as a question generated for the document to be processed. 4. The method of claim 1, wherein Identifying the text type of the document to be processed according to the text structure includes: identifying whether there is a title structure in the text structure of the document to be processed, and the title structure includes a title or a table; Selecting a generation model corresponding to the text type includes: if there is a title structure in the text structure of the document to be processed, then using the structured and semi-structured problem generation models as the generation model corresponding to the text type ; Generating a question for the document to be processed using the selected generation model includes: generating a question for the document to be processed using the structured and semi-structured question generation model. 5. The method according to claim 4, wherein generating a question for the document to be processed using the structured and semi-structured question generation model, comprising: If there is a title in the text structure of the document to be processed, obtain a paraphrase of the attribute related to the title; A question is generated based on the attribute rephrase. 6. The method according to claim 5, characterized in that, obtaining a paraphrase of attributes related to the title, comprising: Obtain the search click presentation log related to the title; Data mining is performed on the search click presentation log to obtain a retelling of attributes related to the title; The attribute repetition is stored in an attribute repetition table. 7. The method of claim 6, wherein generating a question based on the attribute paraphrase comprises: generating a question using the first encoder-decoder model based on the attribute recap; or, Query the attribute paraphrase related to the title from the property paraphrase table, and generate a question according to the property paraphrase found. 8. The method of claim 1, wherein: Identifying the text type of the document to be processed according to the text structure includes: identifying whether there is a question-and-answer structure and a title structure in the text structure of the document to be processed, and the title structure includes a title or a table; Selecting a generation model corresponding to the text type includes: if there is no question-and-answer structure and no title structure in the text structure of the document to be processed, using the natural language question generation model as a generation model corresponding to the text type ; Generating a question for the document to be processed using the selected generation model includes: generating a question for the document to be processed by using the natural language question generation model. 9. The method according to claim 8, wherein generating a question for the document to be processed using the natural language question generation model, comprising: Use the second recurrent neural network model to screen out target sentences from the documents to be processed, where the target sentences include sentences with complete semantics; using a third recurrent neural network model to select candidate answer segments from the target sentence; Based on the candidate answer segments, questions are generated using a second encoder-decoder model. 10 . The method according to claim 1 , further comprising: positioning an answer boundary for the generated question. 11 . 11. The method according to claim 10, wherein positioning the answer boundary for the generated question, comprising: Use the bidirectional attention flow network to predict the start and end positions of the answer segments corresponding to the question; The answer segments are sorted by a learning sorting model, and the answer boundary is located for the question according to the sorting result, wherein the features of the learning sorting model include the start and end positions of the answer segments. 12. A problem generating device, characterized in that, comprising: A text type identification unit, configured to identify the text type of the document to be processed according to the text structure, the text type includes at least one of an explicit FAQ text type, a structured and semi-structured text type, and a natural language text type; a generative model selection unit for selecting a generative model corresponding to the text type, the generative model includes at least one of an explicit question generation model, a structured and semi-structured question generation model, and a natural language question generation model; In the case that the text type is an explicit FAQ text type, the explicit question generation model is used as the generation model corresponding to the text type; in the case that the text type is a structured and semi-structured text type In this case, the structured and semi-structured question generation models are used as the generation models corresponding to the text types; when the text types are natural language text types, the natural language question generation models are used as the generation models corresponding to the text types. The generative model corresponding to the text type described; A question generating unit, configured to generate a question for the document to be processed by using the selected generation model. 13. The apparatus of claim 12, wherein The text type identification unit includes a first identification subunit, and the first identification subunit is used for: identifying whether there is a question-and-answer structure in the text structure of the document to be processed; The generation model selection unit includes a first selection subunit, and the first selection subunit is used for: if there is a question-and-answer structure in the text structure of the document to be processed, the explicit question generation model is used as the The generative model corresponding to the text type; The question generation unit includes a first generation subunit, and the first generation subunit is configured to: generate a question for the document to be processed by using the explicit question generation model. 14. The apparatus according to claim 13, wherein the first generating subunit is further configured to: Judging whether the question part in the question-and-answer structure matches the corresponding answer part, and screening out the part of the text corresponding to the question-and-answer structure that has been successfully matched as candidate text; classifying the screened candidate texts using the first recurrent neural network model to identify explicit questions from the candidate texts; Treat the explicit question as a question generated for the document to be processed. 15. The apparatus of claim 12, wherein The text type identification unit includes a second identification subunit, and the second identification subunit is used to: identify whether there is a title structure in the text structure of the document to be processed, and the title structure includes a title or a table; The generation model selection unit includes a second selection subunit, and the second selection subunit is used for: if there is a title structure in the text structure of the document to be processed, generate a model for the structured and semi-structured problems as a generative model corresponding to the text type; The question generation unit includes a second generation subunit, and the second generation subunit is configured to: generate a question for the document to be processed by using the structured and semi-structured question generation model. 16. The apparatus according to claim 15, wherein the second generating subunit comprises: A paraphrase acquisition subunit, used for acquiring a paraphrase of an attribute related to the title when there is a title in the text structure of the document to be processed; The paraphrase question generation subunit is used to paraphrase the generation question according to the attribute. 17. The apparatus according to claim 16, wherein the paraphrase acquisition subunit is further configured to: Obtain the search click presentation log related to the title; Data mining is performed on the search click presentation log to obtain a retelling of attributes related to the title; The attribute repetition is stored in an attribute repetition table. 18. The apparatus according to claim 17, wherein the paraphrase question generation subunit is further configured to: generating a question using the first encoder-decoder model based on the attribute recap; or, Query the attribute paraphrase related to the title from the property paraphrase table, and generate a question according to the property paraphrase found. 19. The apparatus of claim 12, wherein The text type identification unit includes a third identification subunit, and the third identification subunit is used to: identify whether there is a question-and-answer structure and a title structure in the text structure of the document to be processed, and the title structure includes a title or a table; The generation model selection unit includes a third selection subunit, and the third selection subunit is used for: if there is no question-and-answer structure and no title structure in the text structure of the document to be processed, generate a model for the natural language question as a generative model corresponding to the text type; The question generating unit includes a third generating subunit, and the third generating subunit is configured to: generate a question for the document to be processed by using the natural language question generating model. 20. The apparatus according to claim 19, wherein the third generating subunit is further used for: Use the second recurrent neural network model to screen out target sentences from the documents to be processed, where the target sentences include sentences with complete semantics; using a third recurrent neural network model to select candidate answer segments from the target sentence; Based on the candidate answer segments, questions are generated using a second encoder-decoder model. 21. The apparatus according to any one of claims 12-20, further comprising an answer boundary positioning unit, configured to perform answer boundary positioning for the generated question. 22. The apparatus according to claim 21, wherein the answer boundary positioning unit is further used for: Use the bidirectional attention flow network to predict the start and end positions of the answer segments corresponding to the question; The answer segments are sorted by a learning sorting model, and the answer boundary is located for the question according to the sorting result, wherein the features of the learning sorting model include the start and end positions of the answer segments. 23. A problem generating device, characterized in that, comprising: one or more processors; a storage device for storing one or more programs; The one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method of any of claims 1-11. 24. A computer-readable storage medium storing a computer program, characterized in that, when the program is executed by a processor, the method according to any one of claims 1-11 is implemented.

Images