CN111708870A - Deep neural network-based question answering method and device and storage medium - Google Patents
Deep neural network-based question answering method and device and storage medium Download PDFInfo
- Publication number
- CN111708870A CN111708870A CN202010460124.3A CN202010460124A CN111708870A CN 111708870 A CN111708870 A CN 111708870A CN 202010460124 A CN202010460124 A CN 202010460124A CN 111708870 A CN111708870 A CN 111708870A
- Authority
- CN
- China
- Prior art keywords
- preset
- question
- information
- sentences
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention discloses a question-answering method, a device and a storage medium based on a deep neural network, which are applied to intelligent question-answering equipment, wherein the method comprises the following steps: carrying out text error correction processing on the input problem to obtain an information statement; inputting the information sentences into a preset algorithm library, and outputting reply sentences, wherein the preset algorithm library comprises a task type dialogue algorithm and a retrieval type question-answer algorithm based on a deep neural network, the task type dialogue algorithm is used for identifying task intentions of preset types in the information sentences, and the retrieval type question-answer algorithm based on the deep neural network is used for identifying the similarity between the information sentences and the sentences in the preset question library. The invention can solve the problems of low answer accuracy, long response time, incapability of comprehensively processing the questions and answers in the task and vertical fields at present.
Description
Technical Field
The invention relates to artificial intelligence, in particular to a question answering method and device based on a deep neural network and a storage medium.
Background
The intelligent question-answering system orderly and scientifically arranges the accumulated unordered corpus information and establishes a knowledge-based classification model; the classification models can guide the newly added corpus consultation and service information, save human resources, improve the automation of information processing and reduce the operation cost of the website. The method is characterized in that common questions and answers about the basic conditions of governments and enterprises accumulated for many years on the basis of websites are organized into a standard question-answer library form so as to support intelligent question answering of various types of questions.
However, in the current intelligent question-answering system, the problems of low answer accuracy, long response time and the like generally exist.
Disclosure of Invention
The invention provides a question-answering method, a device and a storage medium based on a deep neural network, aiming at the problems of low answer accuracy and long response time in the prior art.
The technical scheme provided by the invention for the technical problem is as follows:
in a first aspect, the present invention provides a question-answering method based on a deep neural network, where the method includes:
carrying out text error correction processing on the input problem to obtain an information statement;
inputting the information sentences into a preset algorithm library, and outputting reply sentences, wherein the preset algorithm library comprises a task type dialogue algorithm and a retrieval type question-answer algorithm based on a deep neural network, the task type dialogue algorithm is used for identifying task intentions of preset types in the information sentences, and the retrieval type question-answer algorithm based on the deep neural network is used for identifying the similarity between the information sentences and the sentences in the preset question library.
According to the deep neural network-based question answering method, the method comprises the following steps:
the inputting the information statement into a preset algorithm library and outputting a reply statement comprises:
identifying the information statement by using the task type dialogue algorithm;
if the information statement is identified not to contain the task intention of the preset type, identifying the information statement by utilizing the retrieval type question-answering algorithm based on the deep neural network;
and if the similarity between the information sentence and the sentences in the preset question bank is identified to be more than the preset similarity, outputting a reply sentence corresponding to the sentence with the similarity more than the preset similarity.
According to the above question-answering method based on the deep neural network, the inputting of the information sentence into a preset algorithm library and the outputting of a reply sentence further includes:
identifying the information sentences by utilizing the retrieval type question-answering algorithm based on the deep neural network;
if the similarity between the information statement and the statement in a preset question bank is identified to be less than or equal to the preset similarity, identifying the information statement by using the task type dialogue algorithm;
and if the information statement is identified to contain the task intention of the preset type, outputting a reply statement matched with the task intention of the preset type.
According to the above question-answering method based on the deep neural network, the text error correction processing on the input question to obtain the information statement includes:
performing statement preprocessing on the input problem based on a preset rule and a language model to obtain a text character string conforming to the preset rule;
performing a confusability calculation on the text character string by using a confusability dictionary library to determine a wrong word in the text character string;
acquiring the error words, the positions of the error words and the error types; acquiring a candidate word list for replacing the error word;
calculating the confusion degree of the sentence obtained after the candidate words in the candidate word list are used for replacing the error words;
and outputting the sentence with the lowest confusion degree.
According to the above question-answering method based on the deep neural network, the identifying the information statement by using the task-based dialogue algorithm includes:
carrying out a BIEO coding mode on the information statement by using a bidirectional long-time memory model and a conditional random field model to obtain a slot position mark list;
and determining whether the information statement contains a preset type of task intention according to the slot position mark list, wherein the preset type of task intention at least comprises a location query task and/or a combined task intention.
According to the above question-answering method based on the deep neural network, the identifying the information statement by using the retrieval type question-answering algorithm based on the deep neural network comprises:
performing word segmentation processing and word deactivation processing on the information sentences to obtain a corpus result;
recalling a preset number of high-order similar sentences which are arranged according to the similarity level sequence from a preset question bank, wherein the preset number of high-order similar sentences and the corpus results have similarity;
acquiring the depth semantic similarity of the high-order similar sentences of the preset number by using a preset depth model;
rearranging the preset number of high-order similar sentences according to the depth semantic similarity and the literal similarity to obtain similar sentences arranged in the sequence of the comprehensive similarity;
and acquiring similar sentences which are arranged according to the sequence of the comprehensive similarity and have the sentence similarity higher than the preset similarity.
According to the question-answering method based on the deep neural network, the highest-order similar sentences arranged according to the comprehensive similarity in the high-low order are obtained.
According to the deep neural network-based question answering method, the method comprises the following steps:
if the similarity between the information statement and the statement in the preset question bank is identified to be less than or equal to the preset similarity, sensitive word detection is carried out on the information statement;
and if the information sentence is detected to contain the sensitive words, randomly outputting a reply sentence in a preset question-answer result library.
In a second aspect, the present invention further provides a question-answering device based on a deep neural network, which is applied to an intelligent question-answering apparatus, and the device includes:
the preprocessing module is used for carrying out text error correction processing on the input problem to obtain an information statement;
the preprocessing module is used for carrying out text error correction processing on the input problem to obtain an information statement;
the processing module is used for inputting the information sentences into a preset algorithm library and outputting reply sentences, the preset algorithm library comprises a task type dialogue algorithm and a retrieval type question-answer algorithm based on a deep neural network, the task type dialogue algorithm is used for identifying task intentions of preset types in the information sentences, and the retrieval type question-answer algorithm based on the deep neural network is used for identifying the sentence similarity between the information sentences and the preset question library.
In a third aspect, the present invention further provides a question answering storage medium based on the deep neural network, on which a computer program is stored, and the computer program, when executed by a processor, implements the steps of the question answering method based on the deep neural network.
The technical scheme provided by the embodiment of the invention has the following beneficial effects:
the question answering method based on the deep neural network provided by the invention firstly utilizes a text error correction algorithm to process the questions input by the user to obtain information sentences which meet the specifications, then utilizes a task type dialogue algorithm and a retrieval type question answering algorithm based on the deep neural network to respectively identify the information sentences, and returns corresponding question answering results after identification so as to solve the problems that the current answer accuracy is low, the response time is long, and the comprehensive processing of the questions and answers in the task and vertical fields cannot be realized.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a flowchart of a deep neural network-based question answering method according to a first embodiment of the present invention;
FIG. 2 is a schematic diagram of a correction process of the deep neural network-based question answering system according to the present invention;
FIG. 3 is a schematic flow diagram of a FAQbot module of the deep neural network-based question-answering system provided by the present invention;
FIG. 4 is a flowchart of a deep neural network-based question answering method according to a second embodiment of the present invention;
FIG. 5 is a flow chart of a deep neural network-based question answering method according to a third embodiment of the present invention;
FIG. 6 is a schematic structural diagram of an overall module of a deep neural network-based question answering system provided by the present invention;
fig. 7 is a schematic block structure diagram of a deep neural network-based question answering apparatus provided in the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be described in detail with reference to the accompanying drawings.
Referring to fig. 1, a flowchart of a deep neural network-based question answering method according to a first embodiment of the present invention is shown. The question-answering method based on the deep neural network is mainly applied to intelligent question-answering equipment, provides better information service for users through question-answering interaction with the users, and improves the question-answering experience of the users.
Step S101: and carrying out text error correction processing on the input problem to obtain an information statement.
In this step, the input question may be a question in a text form input by the user, or a question obtained by converting a voice input by the user.
With reference to fig. 2, fig. 2 is a schematic diagram of a correction flow of the deep neural network-based question-answering system provided by the present invention. Before text error correction is carried out, the text can be subjected to preliminary processing to obtain a standard sentence, and the preliminary processing can be case conversion, such as converting the problem that the user inputs the English letter related to 'how much IPAD selling price' into 'how much IPAD selling price'. Here, the canonical statement is a statement that conforms to a custom rule that facilitates statement identification.
And performing spelling correction on the errors which may occur, wherein the spelling correction can be performed on the basis of preset rules and a language model, and specifically can include sentence prediction processing, error detection recall processing and correction sequencing processing, the sentence prediction processing can adopt a regular expression to format the character strings to obtain texts in corresponding formats, and operations such as full-angle to half-angle, upper-case to lower-case and the like are performed to convert the sentences into text character strings in lower-case half-angle. The error detection recall processing may calculate, by the confusion degree, a suspected error word in a detection sentence, acquire the suspected error word and a position and an error type thereof, and further acquire a candidate word list that can replace the suspected error word. The correcting and sorting process comprises the following steps: and replacing corresponding suspected error words with candidate words in the candidate word list to obtain sentences, then carrying out sequencing calculation on the sentences to obtain a confusion score list, and finally returning the sentences with the lowest confusion to obtain error correction texts, namely the information sentences.
In a specific application example, for example, inputting the wrong sentence "how many notebook keepers are" which may find the word "keepers" in the vertical domain through error detection, where the word "keepers" may be a wrong word, obtaining a candidate word list [ 'keepers', 'selling price', 'collecting', ] according to the confusion dictionary, sequentially replacing the candidate word with the wrong word "keepers" to calculate the confusion score, taking the candidate word "selling price" with the smallest sentence confusion score, and replacing the wrong word "keepers", so as to obtain the final error-corrected text: "what the price of the notebook is".
It can be understood that the information statement with corresponding requirements can be obtained by adopting a positive and negative case training method and then performing data enhancement on the input problem.
Step S102: inputting the information sentences into a preset algorithm library, and outputting reply sentences, wherein the preset algorithm library comprises a task type dialogue algorithm and a retrieval type question-answer algorithm based on a deep neural network, the task type dialogue algorithm is used for identifying task intentions of preset types in the information sentences, and the retrieval type question-answer algorithm based on the deep neural network is used for identifying the similarity between the information sentences and the sentences in the preset question library.
In this step, the preset type of task intent at least includes a location query task and/or a combined task intent.
The information statement is recognized by a task-based dialog algorithm, i.e. by a task-based dialog system (TaskBot system). The task module of the task-based dialog system may specifically employ a Bi-directional Long Short-Term Memory (Bi) and a conditional random field model (conditional random Fields, CRF) to mark and process the information statements, and specifically, the Bi-directional Long-Term Memory (Bi) and the conditional random field model are employed to identify time, place, task and command word in the statements, including that the statements are subjected to BIEO coding mode marking to obtain a slot marking list, where a tag b (begin) represents the beginning of a segment, a tag i (intermediate) represents the middle of the segment, a tag e (end) represents the end of the segment, and a tag o (other) represents an element unrelated to marking. And determining a location query task or a combined task according to the slot position mark list, and returning slot position information and intention. In addition, the slot marker list may be analyzed and the error markers therein modified to obtain a more accurate slot marker list.
In one example, if the information statement is "charge in the living room after three hours", the slot tag list is obtained through model processing as a combined task with [ 'B-TIM', 'I-TIM', 'E-TIM', 'O', 'B-LOC', 'E-LOC', 'B-ACT', 'E-ACT' ] and intention information as "intetc", the time word is "three hours", the place word is "living room", and the task word is "charge" according to the slot tag list, the slot tag list is analyzed and the error tag return time words, place words, and task slot words are modified, and then the final result is obtained according to the returned words: "combine tasks | three hours; a living room; charge ", directly return to execute the command and wait for the next input of the text sentence. Here, the return execution command may be a reply sentence that establishes a matching relationship with the corresponding task intention in advance.
It can be understood that if the input sentence is a non-task sentence, that is, the input sentence cannot embody the task intention of the user, the intention slot is returned to "other" after the analysis. If the input sentence is "what was eaten in the morning" in one example, when the intention is recognized as "other", that is, a task intention of a non-preset type, it is output as an information sentence.
In this step, the identifying the information statement by using the retrieval type question-answering algorithm based on the deep neural network and the identifying by using the retrieval type question-answering system (FQBot system) may specifically include:
1) performing word segmentation processing and stop word processing on the information sentences to obtain a corpus result;
2) recalling a preset number of high-order similar sentences which are arranged according to the similarity level sequence from a preset question bank, wherein the preset number of high-order similar sentences and the corpus results have similarity;
3) acquiring the depth semantic similarity of the high-order similar sentences of the preset number by using a preset depth model;
4) rearranging the preset number of high-order similar sentences according to the depth semantic similarity and the literal similarity to obtain similar sentences arranged in the sequence of the comprehensive similarity;
5) and acquiring similar sentences which are arranged according to the sequence of the comprehensive similarity and have the sentence similarity higher than the preset similarity.
Fig. 3 is a schematic flow diagram of the FAQBot module of the deep neural network-based question-answering system provided by the invention. The method specifically comprises the following steps:
firstly, the information sentence is cut and stop the word processing to obtain the corpus result, if the input sentence is 'the computer selling price is' the number of the sentence, the corpus result can be 'the computer' and 'the selling price' by the word cutting and stop the word processing.
Secondly, recalling a preset number of high-order similar sentences which are arranged in the similarity degree high-low order from a preset question bank through lucence, for example, the K similar sentences which are ranked at the front can be recalled (namely, the question similar to the 'computer selling price' is), wherein the purpose of recalling the K similar sentences which are ranked at the front comprises obtaining higher question and answer accuracy rate on the basis of improving the recall rate.
And coding the recalled sentences and reordering by using a preset depth model to obtain the depth semantic similarity, wherein the preset depth model is a SimNet model. And then, rearranging the sentences according to the depth semantic similarity and the jcard literal similarity to obtain similar sentences of the K similar sentences arranged according to the comprehensive similarity in high-low order, and outputting question and answer results corresponding to the similar sentences as answers of the input questions if the similar sentences arranged according to the comprehensive similarity in high-low order with the highest order are taken. Here, ranking in the order of the integrated similarity may be achieved by scoring similar sentences.
In one example, if the message statement is "what the computer selling price is", the list (i.e. corpus results) is obtained after cutting and deactivating words: [ 'computer', 'sales price' ], and then enter the lucent recall to get the top K most similar problem lists: ' how much the notebook computer is sold ', ' how much the mail notebook is sold ', ' how much the notebook is taken free ', ' how much the notebook is not to be sold ', ' how much the notebook is mailed ', ' how much the notebook is used ', ' how much the highest similarity score ' how much the notebook computer is sold ' and the final similarity score ' 0.9636 ', and a question-answer result of ' how much the notebook computer is sold ' is returned as a correct question-answer result of the input sentence ' how much the computer is sold ' by comparing the score is above the preset similarity score.
In addition, if the sentence similarity does not reach the similar sentences arranged according to the comprehensive similarity in the high-low order above the preset similarity, the subsequent processing is carried out.
It is understood that the preset depth model can be matched with similar sentences by using depth algorithms such as LSTM, ABCNN, etc.
The question answering method based on the deep neural network provided by the invention firstly utilizes a text error correction algorithm to process the questions input by the user to obtain information sentences which meet the specifications, then utilizes a task type dialogue algorithm and a retrieval type question answering algorithm based on the deep neural network to respectively identify the information sentences, and returns corresponding question answering results after identification so as to solve the problems that the current answer accuracy is low, the response time is long, and the comprehensive processing of the questions and answers in the task and vertical fields cannot be realized.
Referring to fig. 4, a flowchart of a deep neural network-based question answering method according to a second embodiment of the present invention is shown. The present embodiment is different from the first embodiment mainly in that the information sentence is recognized by a task-based dialogue algorithm, and then by a search-based question-and-answer algorithm based on a deep neural network. The implementation mode specifically comprises the following steps:
step S201: and carrying out text error correction processing on the input problem to obtain an information statement.
Step S202: and identifying the information statement by using a task type dialogue algorithm, and outputting a reply statement matched with the task intention of the preset type if the information statement is identified to contain the task intention of the preset type.
Step S203: and if the information statement does not contain the task intention of the preset type, identifying the information statement by utilizing a retrieval type question-answering algorithm based on a deep neural network.
Step S204: and if the similarity between the information sentence and the sentences in the preset question bank is identified to be more than the preset similarity, outputting a reply sentence corresponding to the sentence with the similarity more than the preset similarity.
It is understood that, as a variation of the present embodiment, the steps S202 and S203 can exchange the order with each other, so that after the information sentence does not satisfy the identification requirement in one step, the information sentence is identified in another step, thereby improving the effective reply capability to the information sentence. Specifically, the method may include the steps of:
carrying out text error correction processing on the input problem to obtain an information statement;
identifying the information sentences by utilizing a retrieval type question-answering algorithm based on a deep neural network, and if the similarity between the information sentences and the sentences in a preset question bank is identified to be more than the preset similarity, outputting reply sentences corresponding to the sentences with the similarity more than the preset similarity;
if the similarity between the information statement and the statement in the preset question bank is identified to be less than or equal to the preset similarity, identifying the information statement by using a task type dialogue algorithm;
and if the information statement is identified to contain the task intention of the preset type, outputting a reply statement matched with the task intention of the preset type.
The embodiment can be a more concrete application example of the first embodiment, improve the answer accuracy, shorten the response time, and realize comprehensive processing tasks and vertical domain question answering.
Referring to fig. 5 and 6, fig. 5 is a flowchart of a deep neural network-based question answering method according to another embodiment of the present invention, and fig. 6 is a schematic structural diagram of an overall module of a deep neural network-based question answering system according to the present invention. This embodiment is different from the previous embodiment in that a sensitive word checking step is further included. As shown in fig. 5 and fig. 6, the deep neural network-based question answering method provided in this embodiment includes the following steps:
step S301: and carrying out text error correction processing on the input problem to obtain an information statement.
Step S302: and identifying the information statement by using a task type dialogue algorithm, and outputting a reply statement matched with the task intention of the preset type if the information statement is identified to contain the task intention of the preset type.
Step S303: and if the information statement does not contain the task intention of the preset type, identifying the information statement by utilizing a retrieval type question-answering algorithm based on a deep neural network.
Step S304: and if the information statement does not contain the task intention of the preset type, identifying the information statement by utilizing a retrieval type question-answering algorithm based on a deep neural network.
Step S305: if the similarity between the information statement and the statement in the preset question bank is identified to be less than or equal to the preset similarity, sensitive word detection is carried out on the information statement;
step S306: and if the information sentence is detected to contain the sensitive words, randomly outputting a reply sentence in a preset question-answer result library.
In step S305 and step S306, it is determined whether the user answers or not there is unfriendly content through the sensitive word detecting step. In specific application, whether sensitive words exist in an input text is detected, if the sensitive words are detected, the sensitive words of the text are replaced by 'XX' for representation, and a question and answer result is randomly selected from a specific preset question and answer result base and output. For example, the input sentence "she is easy to toss", no output is generated when the input sentence passes through the front TaskBot module and the front FAQBOt module, the sensitive word detection step firstly judges that the sensitive word "toss" exists in the input sentence, then the "toss" is replaced by the "XX" to obtain "XX o" of her, and a question and answer result "you chat, i can go" is randomly selected from a specific preset question and answer result library to be returned and output.
It is to be understood that in performing sensitive word detection, a DFA algorithm or a word-segmentation-detection algorithm may be employed.
And if the sensitive words are not detected to be contained in the information sentences, outputting chatting answer sentences in the chatting question-answer library to realize chatting dialogue with the user. The chatting question-answer library is a question-answer library which can be established by collecting and sorting question-answer corpora in daily chatting.
On the basis of the previous embodiment, the problem that the current answer accuracy is low, the response time is long, and the task and the question and answer in the vertical field cannot be comprehensively processed is further solved, and the topic guidance and interesting interaction of the user are facilitated.
Fig. 7 is a schematic diagram of a module structure of the deep neural network-based question answering device provided by the present invention. The invention also provides a question answering device based on the deep neural network, which is applied to intelligent question answering equipment, wherein the question answering device 10 comprises:
the preprocessing module 11 is configured to perform text error correction processing on an input question to obtain an information statement;
the first processing module 12 is configured to identify the information statement by using a task-based dialog algorithm, and if it is identified that the information statement includes a task intention of a preset type, output a reply statement matching the task intention of the preset type;
the second processing module 13 is configured to, when it is identified that the information statement does not include a task intention of a preset type, identify the information statement by using a deep neural network-based retrieval type question-answering algorithm;
and the output module 14 is configured to output a reply sentence corresponding to the sentence with the similarity degree higher than the preset similarity degree when it is identified that the similarity degree between the information sentence and the sentences in the preset question bank is higher than the preset similarity degree.
It is understood that, in a specific product, the first processing module 12, the second processing module 13 and the output module 14 may be used as a processing module to perform corresponding processing, so as to input the information sentence into a preset algorithm library, and output a reply sentence, where the preset algorithm library includes a task-based dialog algorithm and a deep neural network-based search-type question-and-answer algorithm, the task-based dialog algorithm is used to identify a task intention of a preset type in the information sentence, and the deep neural network-based search-type question-and-answer algorithm is used to identify a sentence similarity between the information sentence and a preset question-and-answer library.
The corresponding function processing of the information sentences is realized through each module, so that the problems of low answer accuracy, long response time and incapability of comprehensively processing the tasks and the questions and answers in the vertical field at present are solved.
In addition, the invention also provides a question-answering storage medium based on the deep neural network, and a computer program for realizing the question-answering method is stored on the question-answering storage medium.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (10)
1. A question-answering method based on a deep neural network is characterized by comprising the following steps:
carrying out text error correction processing on the input problem to obtain an information statement;
inputting the information sentences into a preset algorithm library, and outputting reply sentences, wherein the preset algorithm library comprises a task type dialogue algorithm and a retrieval type question-answer algorithm based on a deep neural network, the task type dialogue algorithm is used for identifying task intentions of preset types in the information sentences, and the retrieval type question-answer algorithm based on the deep neural network is used for identifying the similarity between the information sentences and the sentences in the preset question library.
2. The deep neural network-based question answering method according to claim 1, wherein the inputting of the information sentences into a preset algorithm library and the outputting of reply sentences comprises:
identifying the information statement by using the task type dialogue algorithm;
if the information statement is identified not to contain the task intention of the preset type, identifying the information statement by utilizing the retrieval type question-answering algorithm based on the deep neural network;
and if the similarity between the information sentence and the sentences in the preset question bank is identified to be more than the preset similarity, outputting a reply sentence corresponding to the sentence with the similarity more than the preset similarity.
3. The deep neural network-based question answering method according to claim 1, wherein the inputting of the information sentence into a preset algorithm library and the outputting of a reply sentence further comprises:
identifying the information sentences by utilizing the retrieval type question-answering algorithm based on the deep neural network;
if the similarity between the information statement and the statement in a preset question bank is identified to be less than or equal to the preset similarity, identifying the information statement by using the task type dialogue algorithm;
and if the information statement is identified to contain the task intention of the preset type, outputting a reply statement matched with the task intention of the preset type.
4. The deep neural network-based question answering method according to any one of claims 1 to 3, wherein the text error correction processing on the input question to obtain the information sentence comprises:
performing statement preprocessing on the input problem based on a preset rule and a language model to obtain a text character string conforming to the preset rule;
performing a confusability calculation on the text character string by using a confusability dictionary library to determine a wrong word in the text character string;
acquiring the error words, the positions of the error words and the error types; acquiring a candidate word list for replacing the error word;
calculating the confusion degree of the sentence obtained after the candidate words in the candidate word list are used for replacing the error words;
and outputting the sentence with the lowest confusion degree.
5. The deep neural network-based question answering method according to any one of claims 2 to 3, wherein the identifying the information sentences by using the task-based dialogue algorithm comprises:
carrying out a BIEO coding mode on the information statement by using a bidirectional long-time memory model and a conditional random field model to obtain a slot position mark list;
and determining whether the information statement contains a preset type of task intention according to the slot position mark list, wherein the preset type of task intention at least comprises a location query task and/or a combined task intention.
6. The deep neural network-based question answering method according to claims 2 to 3, wherein the identifying the information sentences by using the deep neural network-based retrieval type question answering algorithm comprises:
performing word segmentation processing and word deactivation processing on the information sentences to obtain a corpus result;
recalling a preset number of high-order similar sentences which are arranged according to the similarity level sequence from a preset question bank, wherein the preset number of high-order similar sentences and the corpus results have similarity;
acquiring the depth semantic similarity of the high-order similar sentences of the preset number by using a preset depth model;
rearranging the preset number of high-order similar sentences according to the depth semantic similarity and the literal similarity to obtain similar sentences arranged in the sequence of the comprehensive similarity;
and acquiring similar sentences which are arranged according to the sequence of the comprehensive similarity and have the sentence similarity higher than the preset similarity.
7. The deep neural network-based question answering method according to claim 6, wherein the highest-ranking similar sentences arranged in the order of the integrated similarity are obtained.
8. The deep neural network-based question answering method according to claim 1, wherein the method comprises the following steps:
if the similarity between the information statement and the statement in the preset question bank is identified to be less than or equal to the preset similarity, sensitive word detection is carried out on the information statement;
and if the information sentence is detected to contain the sensitive words, randomly outputting a reply sentence in a preset question-answer result library.
9. A question answering apparatus based on a deep neural network, the apparatus comprising:
the preprocessing module is used for carrying out text error correction processing on the input problem to obtain an information statement;
the processing module is used for inputting the information sentences into a preset algorithm library and outputting reply sentences, the preset algorithm library comprises a task type dialogue algorithm and a retrieval type question-answer algorithm based on a deep neural network, the task type dialogue algorithm is used for identifying task intentions of preset types in the information sentences, and the retrieval type question-answer algorithm based on the deep neural network is used for identifying the sentence similarity between the information sentences and the preset question library.
10. A deep neural network-based question answering storage medium, characterized in that a computer program is stored thereon, wherein the computer program, when being executed by a processor, implements the steps of the deep neural network-based question answering method according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010460124.3A CN111708870A (en) | 2020-05-27 | 2020-05-27 | Deep neural network-based question answering method and device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010460124.3A CN111708870A (en) | 2020-05-27 | 2020-05-27 | Deep neural network-based question answering method and device and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111708870A true CN111708870A (en) | 2020-09-25 |
Family
ID=72538033
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010460124.3A Pending CN111708870A (en) | 2020-05-27 | 2020-05-27 | Deep neural network-based question answering method and device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111708870A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112364128A (en) * | 2020-11-06 | 2021-02-12 | 北京乐学帮网络技术有限公司 | Information processing method and device, computer equipment and storage medium |
| CN117034957A (en) * | 2023-06-30 | 2023-11-10 | 海信集团控股股份有限公司 | Semantic understanding method and device |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105468468A (en) * | 2015-12-02 | 2016-04-06 | 北京光年无限科技有限公司 | Data error correction method and apparatus facing question answering system |
| WO2016112679A1 (en) * | 2015-01-14 | 2016-07-21 | 百度在线网络技术（北京）有限公司 | Method, system and storage medium for realizing intelligent answering of questions |
| CN107291828A (en) * | 2017-05-27 | 2017-10-24 | 北京百度网讯科技有限公司 | Spoken inquiry analytic method, device and storage medium based on artificial intelligence |
| CN109410948A (en) * | 2018-09-07 | 2019-03-01 | 北京三快在线科技有限公司 | Communication means, device, system, computer equipment and readable storage medium storing program for executing |
| CN109727041A (en) * | 2018-07-03 | 2019-05-07 | 平安科技（深圳）有限公司 | Intelligent customer service takes turns answering method, equipment, storage medium and device more |
| CN109829044A (en) * | 2018-12-28 | 2019-05-31 | 北京百度网讯科技有限公司 | Dialogue method, device and equipment |
| CN109858007A (en) * | 2017-11-30 | 2019-06-07 | 上海智臻智能网络科技股份有限公司 | Semantic analysis answering method and device, computer equipment and storage medium |
| CN109977208A (en) * | 2019-03-22 | 2019-07-05 | 北京中科汇联科技股份有限公司 | It is a kind of to merge FAQ and task and the actively conversational system of guidance |
| CN110765244A (en) * | 2019-09-18 | 2020-02-07 | 平安科技（深圳）有限公司 | Method and device for acquiring answering, computer equipment and storage medium |
| CN110795548A (en) * | 2019-10-25 | 2020-02-14 | 招商局金融科技有限公司 | Intelligent question answering method, device and computer readable storage medium |
| CN110929016A (en) * | 2019-12-10 | 2020-03-27 | 北京爱医生智慧医疗科技有限公司 | Intelligent question and answer method and device based on knowledge graph |
| CN111078986A (en) * | 2019-12-13 | 2020-04-28 | 腾讯科技（深圳）有限公司 | Data retrieval method, device and computer readable storage medium |
-
2020
- 2020-05-27 CN CN202010460124.3A patent/CN111708870A/en active Pending
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016112679A1 (en) * | 2015-01-14 | 2016-07-21 | 百度在线网络技术（北京）有限公司 | Method, system and storage medium for realizing intelligent answering of questions |
| CN105468468A (en) * | 2015-12-02 | 2016-04-06 | 北京光年无限科技有限公司 | Data error correction method and apparatus facing question answering system |
| CN107291828A (en) * | 2017-05-27 | 2017-10-24 | 北京百度网讯科技有限公司 | Spoken inquiry analytic method, device and storage medium based on artificial intelligence |
| CN109858007A (en) * | 2017-11-30 | 2019-06-07 | 上海智臻智能网络科技股份有限公司 | Semantic analysis answering method and device, computer equipment and storage medium |
| CN109727041A (en) * | 2018-07-03 | 2019-05-07 | 平安科技（深圳）有限公司 | Intelligent customer service takes turns answering method, equipment, storage medium and device more |
| CN109410948A (en) * | 2018-09-07 | 2019-03-01 | 北京三快在线科技有限公司 | Communication means, device, system, computer equipment and readable storage medium storing program for executing |
| CN109829044A (en) * | 2018-12-28 | 2019-05-31 | 北京百度网讯科技有限公司 | Dialogue method, device and equipment |
| CN109977208A (en) * | 2019-03-22 | 2019-07-05 | 北京中科汇联科技股份有限公司 | It is a kind of to merge FAQ and task and the actively conversational system of guidance |
| CN110765244A (en) * | 2019-09-18 | 2020-02-07 | 平安科技（深圳）有限公司 | Method and device for acquiring answering, computer equipment and storage medium |
| CN110795548A (en) * | 2019-10-25 | 2020-02-14 | 招商局金融科技有限公司 | Intelligent question answering method, device and computer readable storage medium |
| CN110929016A (en) * | 2019-12-10 | 2020-03-27 | 北京爱医生智慧医疗科技有限公司 | Intelligent question and answer method and device based on knowledge graph |
| CN111078986A (en) * | 2019-12-13 | 2020-04-28 | 腾讯科技（深圳）有限公司 | Data retrieval method, device and computer readable storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 周鸣争 等: "中国科协新一代信息技术系列丛书 人工智能导论", pages: 165 - 168 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112364128A (en) * | 2020-11-06 | 2021-02-12 | 北京乐学帮网络技术有限公司 | Information processing method and device, computer equipment and storage medium |
| CN117034957A (en) * | 2023-06-30 | 2023-11-10 | 海信集团控股股份有限公司 | Semantic understanding method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108717406B (en) | Text emotion analysis method and device and storage medium | |
| US11693894B2 (en) | Conversation oriented machine-user interaction | |
| CN110795543B (en) | Unstructured data extraction method, device and storage medium based on deep learning | |
| CN110727779A (en) | Question-answering method and system based on multi-model fusion | |
| CN112035730B (en) | Semantic retrieval method and device and electronic equipment | |
| CN110276023B (en) | POI transition event discovery method, device, computing equipment and medium | |
| CN110909122B (en) | Information processing method and related equipment | |
| CN112084381A (en) | Event extraction method, system, storage medium and equipment | |
| CN104471568A (en) | Learning-based processing of natural language questions | |
| CN110096572B (en) | Sample generation method, device and computer readable medium | |
| CN113495900A (en) | Method and device for acquiring structured query language sentences based on natural language | |
| CN114298035A (en) | Text recognition desensitization method and system thereof | |
| CN111782793A (en) | Intelligent customer service processing method, system and equipment | |
| CN111708870A (en) | Deep neural network-based question answering method and device and storage medium | |
| CN113505786A (en) | Test question photographing and judging method and device and electronic equipment | |
| CN112559711A (en) | Synonymous text prompting method and device and electronic equipment | |
| CN110941713B (en) | Self-optimizing financial information block classification method based on topic model | |
| US20120197894A1 (en) | Apparatus and method for processing documents to extract expressions and descriptions | |
| Karpagam et al. | Deep learning approaches for answer selection in question answering system for conversation agents | |
| CN114398482A (en) | Dictionary construction method and device, electronic equipment and storage medium | |
| CN114154503A (en) | Sensitive data type identification method | |
| CN113688633A (en) | Outline determination method and device | |
| CN117291192B (en) | Government affair text semantic understanding analysis method and system | |
| CN116992874B (en) | Text quotation auditing and tracing method, system, device and storage medium | |
| CN114330345B (en) | Named entity recognition method, training method, device, electronic equipment and medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |