Construction of auxiliary decision-making model, decision-making method and system for question-answering fault diagnosis

Abstract

The present invention discloses a construction of an auxiliary decision model for question-answer fault diagnosis, a decision method and a system, the method comprising: preprocessing the dialogue data in a preset corpus to obtain the word vector of the dialogue data; matching the word vectors of each group of questioners and respondents in the dialogue data with a preset dimension to obtain the word vector of the question containing the answer information; calculating the vector similarity between the word vectors of each group of questions containing the answer information and each word vector in the dialogue data of each group of respondents to obtain the corresponding relationship vector, calculating the correlation value between each relationship vector through a neural network model, and sorting the correlation between the answer of the respondent and the question of the questioner based on the correlation value; completing the auxiliary decision of fault diagnosis according to the sorting. The present invention improves the utilization rate of dialogue information and the accuracy of retrieval by analyzing the data information of the corpus and sorting the analysis information by relevance, and can obtain the answer to the question more accurately.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and in particular to the construction of an auxiliary decision-making model, a decision-making method and a system for question-answer fault diagnosis. Background Art

[0002] With the continuous development of power systems in countries around the world, the pressure on equipment maintenance and fault diagnosis is increasing. Faced with the maintenance difficulties of huge power networks, a large number of power suppliers choose to rely on the Internet and use online customer service to provide more convenient remote maintenance guidance. Therefore, many large power companies have accumulated large-scale fault diagnosis and maintenance data with the operation of customer service platforms. If we can effectively use these existing data and automatically diagnose the fault type based on the fault information description, it will greatly reduce the cost of equipment maintenance and further improve the production efficiency of power companies. However, for the accurate judgment of an equipment fault, the interaction between the questioner and the answerer is particularly important. These dialogue sentences are actually interference items for completing fault diagnosis.

[0003] The development of question-answering systems is changing with each passing day. After Turing proposed the Turing test, in 1961, Green et al. implemented an early database-based question-answering system BASEBALL, which can be used to answer natural language questions such as the time, location, and score of a quarter of American baseball games; in 1966, Jaseph Weizenbaum developed the first question-answering system Eliza for psychotherapy, which achieved dialogue and communication with patients through pattern and keyword matching and substitution methods; in 1975, Shank and his colleagues built the SAM system at Yale University, which has the function of reading comprehension, but requires a script describing the corresponding answers to questions; from the 1990s to the present, the research direction of question-answering systems has shifted to question-answer matching based on large-scale document sets, and the research field has also expanded from the initial limited field to the open field, and the research object has expanded from the original fixed corpus to the Internet, but due to the limitations of the technology at the time, the accuracy and precision of the question-answering system for the open domain of the Web were not high, and it could not provide good commercial services.

[0004] As the integration and development of the Internet with various economic and social fields deepens, all walks of life are testing the waters of "Internet +". Power companies keep up with the trend of the times and apply the Internet in various work fields. For example, the use of dialogue robot systems to assist in fault diagnosis is an important application example. After years of operation, the customer service platforms of major power companies have accumulated a large amount of dialogue information about faults, and exchange information is generated through dialogue. Many of these dialogues do not contain answers to solving the fault problems. Since many dialogues do not contain answers to solving the fault problems, the utilization rate of dialogue information and the accuracy of retrieval are reduced. Summary of the invention

[0005] Therefore, the invention provides a question-and-answer fault diagnosis auxiliary decision-making model construction, decision-making method and system, which overcome the defects of low dialogue information utilization and low retrieval accuracy in the prior art.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a method for constructing an auxiliary decision model for question-answer fault diagnosis, comprising:

[0008] Preprocessing the dialogue data in a preset corpus to obtain word vectors of the dialogue data, wherein the dialogue data in the preset corpus includes: dialogue data between at least one group of respondents and a questioner;

[0009] Match the word vectors of each group of questioners and respondents in the conversation data with preset dimensions, adjust the word vectors of the questioners and respondents to word vectors with consistent dimensions, and obtain the word vectors of the questions containing the answer information;

[0010] The word vectors of each group of questions containing answer information are calculated for vector similarity with each word vector in the corresponding group of respondent dialogue data to obtain the corresponding relationship vectors. The correlation values ​​between the relationship vectors are calculated through the neural network model, and the relevance of the respondent's answers to the questioner's questions is ranked based on the correlation values.

[0011] According to the ranking of the relevance between the respondent's answer and the questioner's question, auxiliary decision making for fault diagnosis is completed.

[0012] In one embodiment, the preprocessing of the conversation data includes:

[0013] The conversation data is divided into individual words, and the words with preset parts of speech are removed to obtain the vocabulary of the conversation data. The vocabulary of the conversation data is converted into the word vector of the conversation data through machine learning methods.

[0014] In one embodiment, the preset corpus includes conversation data in multiple languages.

[0015] In one embodiment, the algorithm for sorting the relevance of the respondent's answer to the questioner's question includes: a pointwise sorting algorithm, a pairwise sorting algorithm, and a listwise sorting algorithm.

[0016] In a second aspect, an embodiment of the present invention provides a system for building an auxiliary decision model for question-answer fault diagnosis, including:

[0017] A data processing module, used for preprocessing the dialogue data in a preset corpus to obtain word vectors of the dialogue data, wherein the dialogue data in the preset corpus includes: dialogue data between at least one group of respondents and questioners;

[0018] A sequence matching module is used to match the word vectors of each group of questioners and respondents in the conversation data with a preset dimension, adjust the word vectors of the questioner and the respondent to word vectors with the same dimension, and obtain the word vector of the question containing the answer information;

[0019] The result sorting module is used to calculate the vector similarity between the word vector of each group of questions containing answer information and each word vector in the corresponding group of respondent dialogue data to obtain the corresponding relationship vector, calculate the correlation value between each relationship vector through the neural network model, and sort the relevance of the respondent's answer to the questioner's question based on the correlation value;

[0020] The auxiliary decision-making module is used to complete auxiliary decision-making of fault diagnosis according to the ranking of the relevance between the respondent's answer and the questioner's question.

[0021] In a third aspect, an embodiment of the present invention provides a decision-making assistance method for question-and-answer fault diagnosis, comprising:

[0022] Get the user's request question;

[0023] Preprocess the user's request question to obtain the word vector corresponding to the request question;

[0024] The word vector corresponding to the request question is input into the auxiliary decision module of the auxiliary decision model construction system for question-answer fault diagnosis in the second aspect of the embodiment of the present invention, and the ranking of the relevance of the answers of the respondents in the preset corpus to the user's question is obtained, and the answer to the user's request question is returned according to the ranking result.

[0025] In a fourth aspect, an embodiment of the present invention provides a question-and-answer fault diagnosis auxiliary decision-making system, including:

[0026] A user request question acquisition module is used to obtain the user's request question;

[0027] The request question processing module is used to pre-process the user's request question to obtain the word vector corresponding to the request question;

[0028] The auxiliary decision-making module for the request question inputs the word vector corresponding to the request question into the auxiliary decision-making module of the question-answering fault diagnosis auxiliary decision-making model construction system of the second aspect of the embodiment of the present invention, obtains the ranking of the relevance of the answers of the respondents in the preset corpus to the user's question, and returns the answer to the user's request question according to the ranking result.

[0029] In a fifth aspect, an embodiment of the present invention provides a terminal, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the auxiliary decision-making model construction method for question-and-answer fault diagnosis described in the first aspect of the embodiment of the present invention or the auxiliary decision-making method for question-and-answer fault diagnosis described in the third aspect of the embodiment of the present invention.

[0030] In a sixth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the method for constructing an auxiliary decision-making model for question-and-answer fault diagnosis described in the first aspect of the embodiment of the present invention or the method for auxiliary decision-making for question-and-answer fault diagnosis described in the third aspect of the embodiment of the present invention.

[0031] The technical solution of the present invention has the following advantages:

[0032] The auxiliary decision model construction, decision method and system for question-answer fault diagnosis provided by the present invention preprocesses the dialogue data in a preset corpus to obtain the word vector of the dialogue data, wherein the dialogue data in the preset corpus includes: dialogue data between at least one group of respondents and questioners; matching the word vectors of each group of questioners and respondents in the dialogue data with preset dimensions, adjusting the word vectors of the questioners and respondents to word vectors with consistent dimensions, and obtaining the word vectors of questions containing answer information; calculating the vector similarity between the word vectors of each group of questions containing answer information and each word vector in the dialogue data corresponding to each group of respondents, obtaining the corresponding relationship vectors, calculating the correlation values ​​between the relationship vectors through a neural network model, and ranking the correlation between the answers of the respondents and the questions of the questioners based on the correlation values; completing the auxiliary decision of fault diagnosis according to the ranking of the correlation between the answers of the respondents and the questions of the questioners. The utilization rate of dialogue information and the accuracy of retrieval are improved, and the answers to the questions requested by users can be obtained more accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0034] Figure 1 A flowchart of a specific example of a method for constructing an auxiliary decision-making model for question-answer fault diagnosis provided by an embodiment of the present invention;

[0035] Figure 2 A model diagram of a specific example of the Subtract method in the method for building an auxiliary decision-making model for question-answer fault diagnosis provided by an embodiment of the present invention;

[0036] Figure 3 A model diagram of a specific example of a CNN model in the method for building an auxiliary decision-making model for question-answer fault diagnosis provided by an embodiment of the present invention;

[0037] Figure 4 A module composition diagram of a system for building an auxiliary decision-making model for question-and-answer fault diagnosis provided by an embodiment of the present invention;

[0038] Figure 5 A composition diagram of a specific example of a terminal provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0039] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0041] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, it can also be the internal connection of two components, it can be a wireless connection, or it can be a wired connection. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0042] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0043] Example 1

[0044] A method for constructing an auxiliary decision-making model for question-and-answer fault diagnosis provided by an embodiment of the present invention is as follows Figure 1 shown, and includes the following steps:

[0045] Step S1: Preprocess the dialogue data in the preset corpus to obtain word vectors of the dialogue data. The dialogue data in the preset corpus includes: at least one set of dialogue data between a responder and a questioner.

[0046] In an embodiment of the present invention, the preset corpus includes dialogue data in multiple languages. For example, the corpus includes: Chinese dialogue data, English dialogue data, Chinese-English mixed dialogue, etc. This is only an example and not limited thereto. The corresponding language corpus is selected according to specific requirements.

[0047] In an embodiment of the present invention, preprocessing the dialogue data includes: splitting the dialogue data into individual words, matching them in sentence units, and requiring exactly the same sentences to match each other. In this application, only the answer that can logically answer the question needs to be retrieved, and there is no need for a complete sentence match. Splitting the dialogue data into individual words can extract the core words in the sentence, obtain the required key information, and perform question-and-answer matching with the best choice, so as to judge the user's search intention and content, and display the content that meets the user's needs. Among them, in English writing, spaces are used as natural delimiters between words, while in Chinese, only characters, sentences, and paragraphs can be delimited briefly through obvious delimiters. Therefore, an algorithm is needed to segment Chinese sentences. English word segmentation needs to use the (Natural Language Toolkit, NLTK) natural language processing toolkit. NLTK includes graphical demonstrations and example data, which explain the language processing tasks supported by the toolkit; Chinese needs to use the jieba library, which is a third-party library for Python and is used for Chinese word segmentation.

[0048] In an embodiment of the present invention, after splitting the dialogue data into individual words, it is necessary to remove the words with preset parts of speech. The preset parts of speech refer to the words that appear frequently in the dialogue but have little actual meaning and do not play a role in searching for information, such as: "de", "zai", "he", "jiezhao", etc. This is only an example and not limited thereto. In actual applications, the corresponding parts of speech are selected for processing. Such words appear frequently in the dialogue data and will reduce the search efficiency. Therefore, when searching for answers or processing search requests, in order to save storage space and improve search efficiency, these words or characters, that is, stop words, are automatically ignored to obtain the vocabulary of the dialogue data.

[0049] In the embodiment of the present invention, after removing the words of the preset part of speech and obtaining the words of the dialogue data, the words of the dialogue data are converted into word vectors of the dialogue data by a machine learning method. For example, after converting the words into vectors by word embedding, it is necessary to use a long short-term memory network (Long Short-Term Memory, LSTM) and a recurrent neural network (Gate Recurrent Unit, GRU) to associate each vector so that each word contains context information. The calculation formula is as follows:

[0050]

[0051]

[0052] The calculation formula is similar to the gate function of LSTM, only the input gate is retained, ⊙ represents the element-wise product, W and b are model parameters, and σ is the activation function. Indicates that Transformed into W i Q is a matrix of the same dimension, Q,a is the question and answer vector before encoding, Encoded question and answer vectors.

[0053] Step S2: Match the word vectors of each group of questioners and respondents in the conversation data with a preset dimension, adjust the word vectors of the questioner and the respondent to word vectors with consistent dimensions, and obtain the word vector of the question containing the answer information.

[0054] In an embodiment of the present invention, the word vectors of each group of questioners and respondents in the dialogue data are matched with preset dimensions, and the word vectors of the questioners and respondents are adjusted to word vectors with consistent dimensions, wherein the dimensions can be set by themselves, or the dimensions can be determined based on the smaller number of word vectors of the questioner and the respondent. This is only an example and is not limited to this. In actual applications, the corresponding preset dimensions are selected according to specific needs. In step S1, the dialogue data is converted into word vectors of the dialogue data. In step S2, the key points of the questions and answers are analyzed by unifying the dimensions. For example: an attention matrix is ​​obtained by taking the vector product of the question vector and the answer vector calculated after encoding, and the word vector of the question containing the answer information is calculated by the attention matrix. The calculation formula is as follows:

[0055]

[0056] Among them, the element in the i-th row and j-th column of the attention matrix is ​​e ij , l q is the number of vectors obtained after question segmentation. Assume that the input question word vector is q i , we can get the word vector aq of the question containing the answer informationj .

[0057] Step S3: perform vector similarity calculation on the word vector of each group of questions containing answer information and each word vector in the corresponding group of respondent conversation data to obtain the corresponding relationship vector, calculate the correlation value between each relationship vector through the neural network model, and sort the relevance of the respondent's answer to the questioner's question based on the correlation value.

[0058] In the embodiment of the present invention, the corresponding relationship vector acquisition method includes: Subtract, Multiply, NN, NTN, EusCos and the like, which are only examples and are not limited thereto. In practical applications, the corresponding method is selected according to specific needs, such as Figure 2 As shown in the figure, the model diagram of the Subtract method is used for calculation, and the calculation formula is as follows:

[0059] t j =f(a j ,aq j )=(a j -aq j )⊙(a j -aq j )

[0060] Among them, t j is the corresponding relationship vector, a j For each word vector in the respondent conversation data, aq j is the word vector of the question containing the answer information.

[0061] In this embodiment of the present invention, the calculated relationship vector t j The correlation values ​​between each relationship vector are calculated through the neural network model, such as Figure 3 As shown, q1,q2,...,q n is a set of question word vectors. After attention weighting, a new set of vectors containing question information is obtained: aq1, aq2…aq m , which is consistent with the answer vector a p1 ,a p2 ,...,a pm After comparison, we can get the relationship vector group t p1 ,t p2 ,…,t pm , and it is used as input. After being processed by CNN, the correlation prediction value is obtained. This is only an example and is not limited to this. In practical applications, the corresponding neural network model is selected according to specific needs. The correlation value calculation formula between each relationship vector is as follows:

[0062]

[0063] in, is the correlation value between each relationship vector.

[0064] In an embodiment of the present invention, the algorithm for sorting the relevance of the answer of the respondent to the question of the questioner includes: pointwise sorting algorithm, pairwise sorting algorithm, and listwise sorting algorithm. Among them, in the question-answering retrieval system, it is necessary to sort the answer set A = {a1, a2, ..., an} searched through the question Q, assuming that the question Q and the answer ai have only two relationships: related and unrelated; pointwise sorting algorithm: for each (Q, ai) pair, a judgment is made, and when there are only two relationships, the result is output, label 0 represents irrelevant, and label 1 represents relevant; pairwise sorting algorithm: for each (Q, ai, aj) pair, a sort is performed, assuming that only one of ai and aj is related to Q, the one related to Q is ranked first, and the one irrelevant is ranked behind; listwise sorting algorithm: for each ai in the answer set A, a final result is to make the relevant ai ranked before the irrelevant aj as much as possible.

[0065] Step S4: According to the ranking of the relevance between the respondent's answer and the questioner's question, auxiliary decision making for fault diagnosis is completed.

[0066] In the embodiment of the present invention, the ranking of the relevance between the answers of the respondents and the questions of the questioner is obtained. The answers are ranked according to the magnitude of the relevance to complete the auxiliary decision of the fault diagnosis, wherein the numerical value with a large relevance is ranked first and is used as the answer to the questioner's question to complete the auxiliary decision of the fault diagnosis.

[0067] In this embodiment, the method for constructing an auxiliary decision-making model for question-answer fault diagnosis by experimental comparison and analysis works as follows:

[0068] First, statistical data: the Chinese data sets managezh and businesszh independently developed by the State Grid Corporation of China are used as corpora. Among them, managezh is a data set built by the company using the company's internal daily management online dialogue data. It is currently in version v1.0. Businesszh is a Chinese dialogue data set built based on the company's external business handling. Both contain rich fault diagnosis information. Through processing, the above two corpora have been converted from dialogue corpora to question-answering corpora in the form of one question corresponding to multiple answers. The corpora are composed of three parts: training set, validation set and test set. The specific statistical data of the managezh corpus are as follows:

[0069] question answer relevant irrelevant train 5053 36420 4946 31474 val 1672 12139 1616 10523 test 1659 12142 1610 10532

[0070] The specific statistics of the businesszh corpus are as follows:

[0071] question answer relevant irrelevant train 14639 60718 18894 41824 val 4864 20282 6445 13837 test 4784 20176 6331 13845

[0072] In terms of evaluation indicators, the Mean Reciprocal Rank (MRR) and the Mean Average Precision (MAP) are two indicators commonly used to measure the effectiveness of search algorithms. For the corpus we use, there may be multiple correct answers, so the evaluation indicators we use are the Mean Reciprocal Rank (MAP) and the Mean Reciprocal Rank (MRR).

[0073] Results analysis: First, we used the compare-aggregate with multiple comparison model and selected Multiply as the comparison method. We trained, verified, and tested the two question-answer corpora of managezh and businesszh, respectively. Then, we randomly input the data into the model to determine the relevance between questions and answers. Finally, we sorted all question-answer pairs with the same questions according to the relevance value and determined the performance of the model according to the value of the evaluation index. We used MAP and MRR as the evaluation indicators to evaluate managezh. The experimental results are shown in the following table:

[0074] MAP MRR train 0.994 0.994 val 0.621 0.624 test 0.633 0.634

[0075] The businesszh was evaluated using two evaluation indicators, MAP and MRR. The experimental results are shown in the following table:

[0076] MAP MRR train 0.997 0.998 val 0.754 0.785 test 0.741 0.778

[0077] It can be found that the verification value and test value of MRR and MAP on the data set businesszh are slightly higher than those of managezh. This is because the former contains a higher proportion of answers related to the question than the latter. It can be seen that on the two Chinese data sets, after the model is processed for Chinese preprocessing, the accuracy of completing the question and answer retrieval task based on ranking learning and deep learning algorithms can meet the basic requirements. Both MRR and MAP have achieved ideal target values. Therefore, it can be judged that the auxiliary decision-making model construction model of question and answer fault diagnosis in this application can better assist users in completing fault diagnosis.

[0078] The method for constructing an auxiliary decision-making model for question-answer fault diagnosis provided in an embodiment of the present invention preprocesses the dialogue data in a preset corpus to obtain the word vector of the dialogue data, wherein the dialogue data in the preset corpus includes: dialogue data between at least one group of respondents and questioners; matching the word vectors of each group of questioners and respondents in the dialogue data with a preset dimension, adjusting the word vectors of the questioners and respondents to word vectors with the same dimension, and obtaining the word vectors of the questions containing answer information; calculating the vector similarity between the word vectors of each group of questions containing answer information and each word vector in the dialogue data corresponding to each group of respondents, obtaining the corresponding relationship vectors, calculating the correlation value between each relationship vector through a neural network model, and ranking the relevance of the respondents' answers to the questions of the questioners based on the correlation value; completing the auxiliary decision-making of fault diagnosis according to the ranking of the relevance of the respondents' answers to the questions of the questioners. The utilization rate of dialogue information and the accuracy of retrieval are improved, and the answers to the questions can be obtained more accurately.

[0079] Example 2

[0080] The embodiment of the present invention provides a system for constructing an auxiliary decision model for question-answer fault diagnosis. Figure 4 As shown, including:

[0081] The data processing module 1 is used to preprocess the dialogue data in the preset corpus to obtain word vectors of the dialogue data, wherein the dialogue data in the preset corpus includes: dialogue data between at least one group of respondents and questioners; this module executes the method described in step S1 of embodiment 1, which will not be repeated here.

[0082] The sequence matching module 2 is used to match the word vectors of each group of questioners and respondents in the dialogue data with preset dimensions, adjust the word vectors of the questioners and respondents to word vectors with consistent dimensions, and obtain the word vectors of the questions containing answer information; this module executes the method described in step S2 of Example 1, which will not be repeated here.

[0083] The result sorting module 3 is used to calculate the vector similarity between the word vector of each group of questions containing answer information and each word vector in the corresponding group of respondent dialogue data to obtain the corresponding relationship vector, calculate the correlation value between each relationship vector through the neural network model, and sort the relevance of the respondent's answer to the questioner's question based on the correlation value; this module executes the method described in step S3 of Example 1, which will not be repeated here.

[0084] The auxiliary decision module 4 is used to complete the auxiliary decision of fault diagnosis according to the ranking of the relevance between the respondent's answer and the questioner's question; this module executes the method described in step S4 of embodiment 1, which will not be repeated here.

[0085] The embodiment of the present invention provides a system for constructing an auxiliary decision model for question-answer fault diagnosis, wherein the dialogue data in a preset corpus is preprocessed by a data processing module to obtain word vectors of the dialogue data, wherein the dialogue data in the preset corpus includes: dialogue data between at least one group of respondents and questioners; a sequence matching module, which is used to match the word vectors of each group of questioners and respondents in the dialogue data with preset dimensions, adjust the word vectors of the questioners and respondents to word vectors with consistent dimensions, and obtain word vectors of questions containing answer information; a result sorting module, which is used to calculate the vector similarity between the word vectors of each group of questions containing answer information and each word vector in the corresponding dialogue data of each group of respondents, obtain corresponding relationship vectors, calculate the correlation value between each relationship vector by a neural network model, and sort the relevance of the answer of the respondent to the question of the questioner based on the correlation value; and a decision-making assistance module, which is used to complete the auxiliary decision of fault diagnosis according to the sorting of the relevance of the answer of the respondent to the question of the questioner. The utilization rate of dialogue information and the accuracy of retrieval are improved, and the answer to the question can be obtained more accurately.

[0086] Example 3

[0087] An embodiment of the present invention provides an auxiliary decision-making method for question-and-answer fault diagnosis, comprising: obtaining a user's request question; preprocessing the user's request question to obtain a word vector corresponding to the request question, wherein the preprocessing comprises: dividing the user's request question into separate words, removing words of preset parts of speech, and encoding the words into a word vector corresponding to the request question; inputting the word vector corresponding to the request question into an auxiliary decision-making module of a system for building an auxiliary decision-making model for question-and-answer fault diagnosis in Example 2, obtaining a ranking of the relevance of the answers of respondents in a preset corpus to the user's question, and returning the answer to the user's request question according to the ranking result.

[0088] The embodiment of the present invention provides a decision-making support method for question-and-answer fault diagnosis, which sorts all answers according to their relevance to the user's request question, converts the one-question-one-answer format in the corpus into a one-question-multiple-answer format, and selects the answer with the highest relevance as the fault diagnosis solution, thereby improving the utilization rate of dialogue information and the accuracy of retrieval, and being able to obtain answers to questions more accurately.

[0089] Example 4

[0090] An embodiment of the present invention also provides an auxiliary decision-making system for question-and-answer fault diagnosis, a user request question acquisition module, used to obtain the user's request question; in the request question processing module, the user's request question is preprocessed to obtain a word vector corresponding to the request question, wherein the preprocessing includes: dividing the user's request question into separate words, removing words with preset parts of speech, and encoding the words into a word vector corresponding to the request question; the auxiliary decision-making module for the request question is used to input the word vector corresponding to the request question into the auxiliary decision-making module of the auxiliary decision-making model construction system for question-and-answer fault diagnosis in Example 2, obtain the ranking of the relevance of the respondents' answers in the preset corpus to the user's question, and return the answer to the user's request question according to the ranking result.

[0091] The auxiliary decision-making system for question-and-answer fault diagnosis provided in the embodiment of the present invention sorts all answers according to their relevance to the user's request question, transforms the one-question-one-answer format in the corpus into a one-question-multiple-answer format, and selects the answer with the highest relevance as the fault diagnosis solution, thereby improving the utilization rate of dialogue information and the accuracy of retrieval, and being able to obtain answers to questions more accurately.

[0092] Example 5

[0093] An embodiment of the present invention provides a terminal, such as Figure 5As shown, it includes: at least one processor 401, such as a CPU (Central Processing Unit), at least one communication interface 403, a memory 404, and at least one communication bus 402. Among them, the communication bus 402 is used to realize the connection and communication between these components. Among them, the communication interface 403 may include a display screen (Display), a keyboard (Keyboard), and the optional communication interface 403 may also include a standard wired interface and a wireless interface. The memory 404 may be a high-speed RAM memory (Random Access Memory, volatile random access memory) or a non-volatile memory (non-volatile memory), such as at least one disk storage. The memory 404 may also be at least one storage device located away from the aforementioned processor 401. Among them, the processor 401 can execute the auxiliary decision model construction method for question and answer fault diagnosis in Example 1 or the auxiliary decision method for question and answer fault diagnosis in Example 3. A set of program codes is stored in the memory 404, and the processor 401 calls the program codes stored in the memory 404 to execute the auxiliary decision model construction method of the question-and-answer fault diagnosis in Example 1 or the auxiliary decision method of the question-and-answer fault diagnosis in Example 3. Among them, the communication bus 402 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The communication bus 402 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 Only one line is used to represent it, but it does not mean that there is only one bus or one type of bus. Among them, the memory 404 may include a volatile memory (English: volatile memory), such as a random access memory (English: random-access memory, abbreviated: RAM); the memory may also include a non-volatile memory (English: non-volatile memory), such as a flash memory (English: flash memory), a hard disk drive (English: hard disk drive, abbreviated: HDD) or a solid-state drive (English: solid-state drive, abbreviated: SSD); the memory 404 may also include a combination of the above types of memory. Among them, the processor 401 may be a central processing unit (English: central processing unit, abbreviated: CPU), a network processor (English: network processor, abbreviated: NP) or a combination of CPU and NP.

[0094] Among them, the memory 404 may include a volatile memory (English: volatile memory), such as a random access memory (English: random-access memory, abbreviated: RAM); the memory may also include a non-volatile memory (English: non-volatile memory), such as a flash memory (English: flash memory), a hard disk drive (English: hard disk drive, abbreviated: HDD) or a solid-state drive (English: solid-state drive, abbreviated: SSD); the memory 404 may also include a combination of the above types of memory.

[0095] The processor 401 may be a central processing unit (CPU), a network processor (NP), or a combination of a CPU and a NP.

[0096] The processor 401 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0097] Optionally, the memory 404 is also used to store program instructions. The processor 401 can call the program instructions to implement the auxiliary decision model construction method for question-answer fault diagnosis in Example 1 or the auxiliary decision method for question-answer fault diagnosis in Example 3 of the present application.

[0098] The embodiment of the present invention also provides a computer-readable storage medium, on which computer-executable instructions are stored, and the computer-executable instructions can execute the auxiliary decision-making model construction method for question-answer fault diagnosis in Example 1 or the auxiliary decision-making method for question-answer fault diagnosis in Example 3. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above-mentioned types of memory.

[0099] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived from these are still within the scope of protection of the invention.

Claims

1. A method for constructing an auxiliary decision-making model for question-answer fault diagnosis, characterized in that: include: Preprocessing the dialogue data in a preset corpus to obtain word vectors of the dialogue data, wherein the dialogue data in the preset corpus includes: dialogue data between at least one group of respondents and questioners, and preprocessing the dialogue data includes: Divide the conversation data into individual words, remove the words with preset parts of speech, obtain the words of the conversation data, and convert the words of the conversation data into word vectors of the conversation data by machine learning methods. The preset parts of speech refer to words that appear frequently in the conversation but have little practical significance and do not play a role in searching for information. The word vectors of each group of questioners and respondents in the dialogue data are matched with preset dimensions, and the word vectors of the questioners and respondents are adjusted to word vectors with the same dimension to obtain the word vectors of the questions containing the answer information; wherein, an attention matrix is ​​obtained by the vector product of the question vector and the answer vector calculated after encoding, and the word vector of the question containing the answer information is calculated by the attention matrix, and the calculation formula is as follows: Among them, the element in the i-th row and j-th column of the attention matrix is ​​e ij , l q is the number of vectors obtained after question segmentation. Assume that the input question word vector is q i , we can get the word vector aq of the question containing the answer information j ; The word vectors of each group of questions containing answer information are calculated for vector similarity with each word vector in the corresponding group of respondent dialogue data to obtain the corresponding relationship vectors. The correlation values ​​between the relationship vectors are calculated through the neural network model, and the relevance between the respondent's answer and the questioner's question is ranked based on the correlation value. The corresponding relationship vector is obtained using the model diagram of the Subtract method, and the calculation formula is as follows: t j =f(a j ,aq j )=(a j -aq j )⊙(a j -aq j ) Among them, t j is the corresponding relationship vector, a j For each word vector in the respondent conversation data, aq j is the word vector of the question containing the answer information; The calculation of the correlation value includes the following steps, where q1, q2, ..., q n is a set of question word vectors. After the attention weighting of the above calculation formula, a new set of vectors containing question information is obtained, aq1, aq2…aq m , which is consistent with the answer vector a p1 ,a p2 ,...,a pm The above calculation formula is used to calculate the relationship vector group t p1 ,t p2 ,…,t pm , taking the relationship vector group as input and processing it through the neural network model to obtain the relevant value; According to the ranking of the relevance between the respondent's answer and the questioner's question, auxiliary decision making for fault diagnosis is completed.

2. The method for constructing an auxiliary decision-making model for question-answer fault diagnosis according to claim 1, characterized in that: The preset corpus includes conversation data in multiple languages.

3. The method for constructing an auxiliary decision-making model for question-answer fault diagnosis according to claim 1, characterized in that: The algorithm for sorting the relevance between the respondent's answer and the questioner's question includes: a pointwise sorting algorithm, a pairwise sorting algorithm, and a listwise sorting algorithm.

4. A system for building a decision-making support model for question-answer fault diagnosis, characterized in that: include: The data processing module is used to pre-process the dialogue data in the preset corpus to obtain word vectors of the dialogue data, wherein the dialogue data in the preset corpus includes: dialogue data between at least one group of respondents and questioners, and the pre-processing of the dialogue data includes: Divide the conversation data into individual words, remove the words with preset parts of speech, obtain the words of the conversation data, and convert the words of the conversation data into word vectors of the conversation data by machine learning methods. The preset parts of speech refer to words that appear frequently in the conversation but have little practical significance and do not play a role in searching for information. The sequence matching module is used to match the word vectors of each group of questioners and respondents in the dialogue data with preset dimensions, adjust the word vectors of the questioners and respondents to word vectors with consistent dimensions, and obtain the word vectors of the questions containing answer information; wherein, an attention matrix is ​​obtained by the vector product of the question vector and the answer vector calculated after encoding, and the word vector of the question containing answer information is calculated by the attention matrix. The calculation formula is as follows: Among them, the element in the i-th row and j-th column of the attention matrix is ​​e ij , l q is the number of vectors obtained after question segmentation. Assume that the input question word vector is q i , we can get the word vector aq of the question containing the answer information j ; The result sorting module is used to calculate the vector similarity between the word vector of each group of questions containing answer information and each word vector in the corresponding group of respondent dialogue data to obtain the corresponding relationship vector, calculate the correlation value between each relationship vector through the neural network model, and sort the relevance of the respondent's answer to the questioner's question based on the correlation value; the corresponding relationship vector is obtained by using the model diagram of the Subtract method, and the calculation formula is as follows: t j =f(a j ,aq j )=(a j -aq j )⊙(a j -aq j ) Among them, t j is the corresponding relationship vector, a j For each word vector in the respondent conversation data, aq j is the word vector of the question containing the answer information; The calculation of the correlation value includes the following steps, where q1, q2, ..., q n is a set of question word vectors. After the attention weighting of the above calculation formula, a new set of vectors containing question information is obtained, aq1, aq2…aq m , which is consistent with the answer vector a p1 ,a p2 ,...,a pm The above calculation formula is used to calculate the relationship vector group t p1 ,t p2 ,…,t pm , taking the relationship vector group as input and processing it through the neural network model to obtain the relevant value; The auxiliary decision-making module is used to complete auxiliary decision-making of fault diagnosis according to the ranking of the relevance between the respondent's answer and the questioner's question.

5. A question-and-answer fault diagnosis decision-making method, characterized in that: include: Get the user's request question; Preprocess the user's request question to obtain the word vector corresponding to the request question; The word vector corresponding to the request question is input into the auxiliary decision module of the auxiliary decision model construction system for question-answer fault diagnosis in claim 4, and the relevance of the answers of the respondents in the preset corpus to the user's question is ranked, and the answer to the user's request question is returned according to the ranking result.

6. A question-answer fault diagnosis auxiliary decision system, characterized in that: include: A user request question acquisition module is used to obtain the user's request question; The request question processing module is used to pre-process the user's request question to obtain the word vector corresponding to the request question; The auxiliary decision-making module for the request question inputs the word vector corresponding to the request question into the auxiliary decision-making module of the auxiliary decision-making model construction system for question-answer fault diagnosis of claim 4, obtains the ranking of the relevance of the answers of the respondents in the preset corpus to the user's question, and returns the answer to the user's request question according to the ranking result.

7. A terminal, characterized in that: include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the auxiliary decision-making model construction method for question-and-answer fault diagnosis as described in any one of claims 1-3 or the auxiliary decision-making method for question-and-answer fault diagnosis as described in claim 5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the auxiliary decision-making model construction method for question-and-answer fault diagnosis described in any one of claims 1-3 or the auxiliary decision-making method for question-and-answer fault diagnosis described in claim 5.

Images