CN109492085B - Answer determination method, device, terminal and storage medium based on data processing - Google Patents
Answer determination method, device, terminal and storage medium based on data processing Download PDFInfo
- Publication number
- CN109492085B CN109492085B CN201811364713.0A CN201811364713A CN109492085B CN 109492085 B CN109492085 B CN 109492085B CN 201811364713 A CN201811364713 A CN 201811364713A CN 109492085 B CN109492085 B CN 109492085B
- Authority
- CN
- China
- Prior art keywords
- answer
- candidate
- preset
- question
- matching degree
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 54
- 238000012545 processing Methods 0.000 title claims abstract description 35
- 238000004364 calculation method Methods 0.000 claims abstract description 17
- 238000012549 training Methods 0.000 claims description 96
- 238000005457 optimization Methods 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 11
- 238000004422 calculation algorithm Methods 0.000 description 11
- 230000006870 function Effects 0.000 description 11
- 238000010586 diagram Methods 0.000 description 7
- 239000013598 vector Substances 0.000 description 7
- 241000282414 Homo sapiens Species 0.000 description 5
- 230000003993 interaction Effects 0.000 description 4
- 230000002457 bidirectional effect Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 230000002787 reinforcement Effects 0.000 description 2
- 238000010845 search algorithm Methods 0.000 description 2
- 241000008357 Okapia johnstoni Species 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the invention discloses an answer determining method, device, terminal and storage medium based on data processing. The method comprises the following steps: acquiring an initial question input by a user, and calling a preset retrieval model to determine a candidate answer set corresponding to the initial question from a preset knowledge base; invoking a preset generation model to determine a generated answer set corresponding to the initial question; calculating candidate matching degrees between each candidate answer in the candidate answer set and the initial question according to a preset calculation rule to obtain at least one candidate matching degree, and solving a first average value of the at least one candidate matching degree; and determining a target answer to be output from the candidate answer set or the generated answer set according to a preset determination rule and the first average value. The embodiment of the invention can better determine the target answer, avoid the long tail problem of the target answer and ensure the consistency and rationality of the target answer.
Description
Technical Field
The present invention relates to the field of computer technologies, and in particular, to a method, an apparatus, a terminal, and a storage medium for determining answers based on data processing.
Background
Human-computer interaction (HCI) refers to determining the information exchange process between a person and a computer in a certain interaction manner by using a certain dialogue language between the person and the computer. With the development of man-machine interaction technology, more and more intelligent products based on man-machine interaction technology, such as chat robots, etc., are generated. The intelligent products can chat with users and generate corresponding answer information according to the questions of the users. But currently, the answer information retrieved by intelligent products according to the questions of users usually has long-tail questions (i.e. minor questions), or consistency and rationality of the answer information are difficult to ensure. Thus, how to better determine the target answer from the user's question becomes a research hotspot.
Disclosure of Invention
The embodiment of the invention provides an answer determining method, device, terminal and computer readable storage medium based on data processing, which can better determine a target answer, avoid long-tail problems of the target answer and ensure consistency and rationality of the target answer.
In one aspect, an embodiment of the present invention provides a data processing-based answer determination method, including:
Acquiring an initial question input by a user, and calling a preset retrieval model to determine a candidate answer set corresponding to the initial question from a preset knowledge base, wherein the preset knowledge base comprises at least one question and one or more answers corresponding to each question, and the candidate answer set comprises at least one candidate answer;
Invoking a preset generation model to determine a generation answer set corresponding to the initial question, wherein the generation answer set comprises at least one generation answer, and the preset generation model is obtained by performing model training optimization by adopting a plurality of training data sets containing the question;
Calculating candidate matching degrees between each candidate answer in the candidate answer set and the initial question according to a preset calculation rule to obtain at least one candidate matching degree, and solving a first average value of the at least one candidate matching degree;
and determining a target answer to be output from the candidate answer set or the generated answer set according to a preset determination rule and the first average value.
In another aspect, an embodiment of the present invention provides a data processing-based answer determining apparatus, including:
The system comprises an acquisition unit, a search unit and a search unit, wherein the acquisition unit is used for acquiring an initial question input by a user, and invoking a preset search model to determine a candidate answer set corresponding to the initial question from a preset knowledge base, the preset knowledge base comprises at least one question and one or more answers corresponding to each question, and the candidate answer set comprises at least one candidate answer;
The acquisition unit is used for calling a preset generation model to determine a generation answer set corresponding to the initial question, wherein the generation answer set comprises at least one generation answer, and the preset generation model is obtained by performing model training optimization by adopting a plurality of training data sets containing the question;
the computing unit is used for respectively computing candidate matching degrees between each candidate answer in the candidate answer set and the initial question according to a preset computing rule so as to obtain at least one candidate matching degree, and solving a first average value of the at least one candidate matching degree;
and the determining unit is used for determining a target answer to be output from the candidate answer set or the generated answer set according to the first average value according to a preset determining rule.
In yet another aspect, an embodiment of the present invention provides a terminal, including an input device, an output device, a memory, and a processor, where the processor, the input device, the output device, and the memory are connected to each other, where the memory is configured to store a computer program, the computer program includes program instructions, and the processor is configured to call the program instructions to perform the steps of:
Acquiring an initial question input by a user, and calling a preset retrieval model to determine a candidate answer set corresponding to the initial question from a preset knowledge base, wherein the preset knowledge base comprises at least one question and one or more answers corresponding to each question, and the candidate answer set comprises at least one candidate answer;
Invoking a preset generation model to determine a generation answer set corresponding to the initial question, wherein the generation answer set comprises at least one generation answer, and the preset generation model is obtained by performing model training optimization by adopting a plurality of training data sets containing the question;
Calculating candidate matching degrees between each candidate answer in the candidate answer set and the initial question according to a preset calculation rule to obtain at least one candidate matching degree, and solving a first average value of the at least one candidate matching degree;
and determining a target answer to be output from the candidate answer set or the generated answer set according to a preset determination rule and the first average value.
In yet another aspect, embodiments of the present invention provide a computer-readable storage medium storing a computer program. The computer program comprises at least one program instruction loadable by a processor and adapted to perform the steps of:
Acquiring an initial question input by a user, and calling a preset retrieval model to determine a candidate answer set corresponding to the initial question from a preset knowledge base, wherein the preset knowledge base comprises at least one question and one or more answers corresponding to each question, and the candidate answer set comprises at least one candidate answer;
Invoking a preset generation model to determine a generation answer set corresponding to the initial question, wherein the generation answer set comprises at least one generation answer, and the preset generation model is obtained by performing model training optimization by adopting a plurality of training data sets containing the question;
Calculating candidate matching degrees between each candidate answer in the candidate answer set and the initial question according to a preset calculation rule to obtain at least one candidate matching degree, and solving a first average value of the at least one candidate matching degree;
and determining a target answer to be output from the candidate answer set or the generated answer set according to a preset determination rule and the first average value.
In the embodiment of the invention, after the initial question input by the user is acquired, a preset retrieval model can be called to determine a candidate answer set corresponding to the initial question from a preset knowledge base, and a preset generation model is called to determine a generation answer set corresponding to the initial question. And then, calculating the candidate matching degree between each candidate answer in the candidate answer set and the initial question respectively to obtain at least one candidate matching degree, and solving a first average value of the at least one candidate matching degree. Finally, the target answer to be output can be determined from the candidate answer set or the generated answer set according to the first average value. The embodiment of the invention invokes the retrieval model and the generation model to determine the target answer, can avoid the long tail problem of the target answer and ensures the consistency and the rationality of the target answer. And determining whether to determine the target answer from the candidate answer set or the generated answer set according to the first average value can avoid the situation of wrong retrieval of the retrieval model, thereby improving accuracy.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following description will simply refer to the drawings that are required to be used in the description of the embodiments of the present invention.
FIG. 1 is a flowchart of an answer determining method based on data processing according to an embodiment of the present invention;
FIG. 2 is a flowchart of an answer determining method based on data processing according to another embodiment of the present invention;
FIG. 3a is an application scenario diagram of an answer determination method based on data processing according to an embodiment of the present invention;
FIG. 3b is a diagram of an application scenario of an answer determination method based on data processing according to an embodiment of the present invention;
Fig. 4 is a schematic structural diagram of an answer determining apparatus based on data processing according to an embodiment of the present invention;
Fig. 5 is a schematic structural diagram of a terminal according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention are described below with reference to the accompanying drawings in the embodiments of the present invention.
The embodiment of the invention provides a data processing-based answer determination method, which can be applied to chat conversations between a terminal and a user, wherein the terminal can comprise, but is not limited to: smart devices such as smartphones, laptops, tablets, desktop computers, and chat devices based on chat conversations, such as chat robots, and the like. Specifically, in the process of chat conversation with the user, the terminal can acquire an initial question input by the user on the user interface, then call a preset retrieval model to determine a candidate answer set corresponding to the initial question from a preset knowledge base, and call a preset generation model to determine a generated answer set corresponding to the initial question. And calculating the candidate matching degree between each candidate answer in the candidate answer set and the initial question according to a preset calculation rule respectively to obtain at least one candidate matching degree, and solving a first average value of the at least one candidate matching degree. And finally, determining the target answer to be output from the candidate answer set or the generated answer set according to a preset determination rule and the first average value. After determining the target answer, the target answer may be output in a user interface to enable a chat session with the user.
Fig. 1 is a flowchart of an answer determining method based on data processing according to an embodiment of the present invention, where the answer determining method may be performed by the above-mentioned terminal. As shown in fig. 1, the answer determining method based on data processing may include the following steps S101 to S104:
S101, acquiring an initial question input by a user, and calling a preset retrieval model to determine a candidate answer set corresponding to the initial question from a preset knowledge base. The preset knowledge base comprises at least one question and one or more answers corresponding to each question, and the candidate answer set comprises at least one candidate answer.
Specifically, when acquiring an initial problem input by a user, the terminal can acquire voice information of the user, and extract the initial problem from the voice information; for example, say "hello, you know what are the components of the computer? The terminal can acquire the voice information and extract the initial problem of which components of the computer are from the voice information. In one embodiment, the terminal may also obtain text information input by the user, and extract an initial question from the text information; for example, the terminal may provide a dialog interface for the user so that the user may enter the text information "hello, is you aware of which of the computer's components are? The terminal can detect the input operation of the user, acquire text information input by the user, and extract an initial problem of which components of the computer are from the text information.
After the initial question is acquired, the terminal may invoke a preset search model to determine a candidate answer set corresponding to the initial question from a preset knowledge base, where the preset search model may be trained based on a preset search algorithm, and the preset search algorithm may include, but is not limited to: IR (Information Retrieval) algorithm, BM25 (Okapi BM 25) algorithm, and so on.
S102, a preset generation model is called to determine a generation answer set corresponding to the initial question, wherein the generation answer set comprises at least one generation answer.
Specifically, the preset generation model is obtained by performing model training optimization by adopting a plurality of training data sets containing problems; in one embodiment, the preset generative model may be a attention +seq2seq based generative model. After acquiring the initial question, the terminal may invoke the preset generating model to generate one or more generating answers for the initial question, and then use a set formed by all generating answers generated by the preset generating model or a preset number of generating answers as a generating answer set.
S103, calculating the candidate matching degree between each candidate answer in the candidate answer set and the initial question according to a preset calculation rule respectively to obtain at least one candidate matching degree, and solving a first average value of the at least one candidate matching degree.
In one embodiment, the preset calculation rule may include: and determining a rule of candidate matching degree according to the frequency of combined occurrence of each candidate answer and the initial question answer. Specifically, based on a preset knowledge base, the frequency of joint occurrence of each candidate answer and the initial question can be counted, and the counted frequency is used as the candidate matching degree between the candidate answers and the initial question, so that at least one candidate matching degree can be obtained. Then, a first average of the at least one candidate matching degree is obtained.
In yet another embodiment, the preset calculation rule may include: and scoring each candidate answer to determine a rule of candidate matching degree. Specifically, a multi-feature statistical machine learning method may be adopted to score each candidate answer for the initial question, so as to obtain a score of each candidate answer, and the score of each candidate answer is used as a candidate matching degree between each candidate answer and the initial question, and then a first average value of at least one candidate matching degree is obtained. Specifically, when the scoring processing is performed on each candidate answer by using a multi-feature statistical machine learning method, the scoring processing may be performed on a plurality of features such as sentence features, language features, vocabulary pattern features, redundancy features, and the like.
It should be noted that, if only one candidate answer is included in the candidate answer set, the first average value is equal to the candidate matching degree between the candidate answer and the initial question.
S104, determining a target answer to be output from the candidate answer set or the generated answer set according to a preset determination rule and the first average value.
After the first average value is obtained, whether to determine the target answer from the candidate answer set or the target answer from the generated answer set may be determined according to a preset determination rule based on the first average value. In one embodiment, the preset determination rules include: determining the target answer according to the magnitude relation between the first average value and the preset threshold value, and correspondingly, determining the target answer to be output from the candidate answer set or the generated answer set according to the preset determination rule according to the first average value may be: judging whether the first average value is larger than a preset threshold value or not, wherein the preset threshold value can be set according to actual service requirements, for example, the preset threshold value can be 0.5; if the first average value is larger than a preset threshold value, selecting a candidate answer with highest candidate matching degree from the candidate answer set as a target answer; if the first average value is not greater than the preset threshold value, selecting the generated answer with the highest generated matching degree from the generated answer set as a target answer, wherein the generated matching degree is the matching degree between the generated answer in the generated answer set and the initial question.
Practice shows that only one candidate answer in the candidate answer set has high matching degree with the initial question, and other candidate answers have low matching degree with the initial question, and the candidate answer with high matching degree is probably searched by mistake due to search errors of the search model. That is, each candidate answer in the candidate answer set in this case is inaccurate, and if the candidate answer with higher matching degree is directly output as the target answer at this time, the accuracy of the target answer is reduced. Therefore, the embodiment of the invention adopts the method of the first average value, determines whether to determine the target answer from the candidate answer set or determine the target answer from the generated answer set according to the magnitude relation between the first average value and the preset threshold value, and can avoid the situation of wrong retrieval of the retrieval model to a certain extent, thereby improving the accuracy of the target answer.
In yet another embodiment, the preset determination rule includes: the determining the target answer according to the magnitude relation between the first average value and the second average value, and correspondingly, according to a preset determining rule, determining the target answer to be output from the candidate answer set or the generated answer set according to the first average value may be: calculating the generation matching degree between each generation answer in the generation answer set and the initial question respectively to obtain at least one generation matching degree, and solving a second average value of the at least one generation matching degree; if the first average value is larger than the second average value, selecting a candidate answer with highest candidate matching degree from the candidate answer set as a target answer; if the first average value is smaller than the second average value, selecting a generated answer with highest generated matching degree from the generated answer set as a target answer; and if the first average value is equal to the second average value, selecting the candidate answer with the highest candidate matching degree from the candidate answer set as a target answer, or selecting the generated answer with the highest generated matching degree from the generated answer set as the target answer.
It should be noted that, if only one generated answer is included in the generated answer set, the second average value is equal to the generated matching degree between the generated answer and the initial question. Therefore, according to the embodiment, whether the target answer is determined from the candidate answer set or the generated answer set is determined according to the magnitude relation between the first average value and the second average value, the situation that the accuracy is reduced due to unreasonable preset threshold value values can be avoided, and the accuracy of the target answer can be further improved.
In the embodiment of the invention, after the initial question input by the user is acquired, a preset retrieval model can be called to determine a candidate answer set corresponding to the initial question from a preset knowledge base, and a preset generation model is called to determine a generation answer set corresponding to the initial question. And then, calculating the candidate matching degree between each candidate answer in the candidate answer set and the initial question respectively to obtain at least one candidate matching degree, and solving a first average value of the at least one candidate matching degree. Finally, the target answer to be output can be determined from the candidate answer set or the generated answer set according to the first average value. The embodiment of the invention invokes the retrieval model and the generation model to determine the target answer, can avoid the long tail problem of the target answer and ensures the consistency and the rationality of the target answer. And determining whether to determine the target answer from the candidate answer set or the generated answer set according to the first average value can avoid the situation of wrong retrieval of the retrieval model, thereby improving accuracy.
Fig. 2 is a flowchart of another answer determining method based on data processing according to an embodiment of the invention, where the answer determining method based on data processing may be performed by the terminal. As shown in fig. 2, the answer determining method based on data processing may include the following steps S201 to S207:
S201, constructing a first training data set, wherein the first training data set comprises at least one pair of chat question-answer corpora, the at least one pair of chat question-answer corpora is obtained from at least one question-answer system, and each pair of chat question-answer corpora comprises a question and a corresponding standard answer.
Specifically, one or more question-answering systems may be searched in advance on the internet, and one or more pairs of chat question-answering corpora are obtained from the one or more question-answering systems, where each pair of chat question-answering corpora is a real and actually existing chat question-answering corpus, and the real and actually existing chat question-answering corpus refers to a corpus composed of questions that have been input by a user in the question-answering system and standard answers output by the question-answering system for the questions. For example, the user inputs "where the hometown is in" in a question and answer system, and the standard answer output by the question and answer system is "Beijing", so that the "where the hometown is in" and "Beijing" can be used as the chat question and answer corpus that is real and actually exists.
S202, acquiring a pre-built initial model, and training the initial model by adopting a first training data set to obtain an intermediate model.
Specifically, the pre-built initial model may include an encoder model and a decoder model, and when the initial model is pre-built, bi-GRU bidirectional GRU models are selected as the encoder model and the decoder model, where the Bi-GRU bidirectional GRU model is a model capable of identifying a flip-chip structure. Because the initial question is likely to be a flip-chip sentence structure when the user inputs the initial question, i.e. different from the normal sentence structure, for example, the initial question input by the user is "where to today", the normal sentence structure is "where to today", and the Bi-GRU bidirectional GRU model is adopted to identify the initial question of the flip-chip sentence structure, thereby enriching the functions of the preset generation model and improving the robustness of the preset generation model.
It should be noted that, in the embodiment of the present invention, bi-GRU Bi-directional GRU models are selected as the encoder model and the decoder model, but the architectures of the encoder model and the decoder model are inconsistent, that is, the model parameters in the encoder model and the decoder model are inconsistent. Because the model parameters of the encoder model and the decoder model are inconsistent, more model parameters need to be trained and updated in the subsequent training process of the generated model, so that the robustness and performance of the generated model can be improved, and the generated answer determined by the preset generated model obtained by training is closer to the language of human beings and has more authenticity.
The first training data set may be input into the initial model while training the initial model with the first training data set. After receiving the first training data set, the encoder model in the initial model may encode the questions in each pair of chat question-answer corpora in the first training data set into feature vectors, and then the decoder model performs decoding processing according to the feature vectors to determine corresponding answers corresponding to the questions. Then judging whether the corresponding answer is consistent with the standard answer corresponding to the question in the first training data set: if the answer is inconsistent, model parameters of the encoder model and the decoder model in the initial model are continuously updated until corresponding answers corresponding to the questions determined by the updated initial model can be consistent with standard answers, and the updated initial model can be used as an intermediate model at the moment; if the answer is consistent, the initial model can accurately determine the corresponding answer corresponding to the question, and the initial model can be directly used as an intermediate model.
S203, acquiring a second training data set according to the first training data set, and adopting the second training data set to optimize the intermediate model to obtain a preset generation model.
After the intermediate model is obtained, a second training data set may be obtained from the intermediate model and the first training data set. In particular, a specific embodiment of acquiring the second training data set according to the first training data set may be: for each question in the first training data set, sequentially calling an intermediate model to determine a corresponding answer corresponding to each question, and taking the corresponding answers corresponding to all questions in the first training data set as negative samples; obtaining standard answers corresponding to each answer in the first training data set, and taking the standard answers corresponding to all questions in the first training data set as positive samples; the data set comprising negative and positive samples is taken as a second training data set. That is, the negative sample included in the second training data set is a corresponding answer generated by the intermediate model, the positive sample is a standard answer corresponding to the real language of the human being, and the negative sample and the positive sample may be collectively referred to as a sample. For example, the problem in the first training data set is "what the component of the computer is", the intermediate model is called to determine that the corresponding answer corresponding to the problem is "① screen, ② keyboard, ③ video card, ④ sound card, ⑤ hard disk", and then the negative sample is "① screen, ② keyboard, ③ video card, ④ sound card, ⑤ hard disk"; the standard answer corresponding to the question in the first training data set is that the computer is generally composed of components such as a screen, a keyboard, a display card, a sound card and a hard disk, and then the positive sample is that the computer is generally composed of components such as a screen, a keyboard, a display card, a sound card and a hard disk.
After the second training data set is obtained, the second training data set can be adopted to perform optimization processing on the intermediate model, so that a preset generation model is obtained. In particular, during the optimization of the intermediate model with the second training data set, a discriminant model may be introduced, which may be a two-class model, with the objective of discriminating whether each sample in the second training data set is a positive or a negative sample. For any sample in the second training data set, a discriminator model can be called to discriminate the sample, so as to obtain the probability that the sample is a positive sample. In the distinguishing process, the distinguishing between the positive sample and the negative sample can be learned by the distinguishing model, and the model parameters of the distinguishing model are updated according to the learned distinguishing, so that the distinguishing capability of the distinguishing model is improved.
After invoking the discriminant model to obtain a probability that the sample is a positive sample, the probability may be used as a reward function (rewards) for the intermediate model, where a larger value of rewards indicates that the sample is more likely to be a positive sample, i.e., the sample is more closely related to human real language. Then, a reinforcement learning algorithm (Policy Gradient algorithm) may be combined, the value of the loss function of the intermediate model is calculated according to rewards, and then the model parameters of the intermediate model are optimized according to the principle of reducing the calculated value of the loss function. And then generating a new corresponding answer again according to the question in the second training data set by adopting the intermediate model after optimizing the parameters, and taking the new corresponding answer as a new negative sample. And then, calling the discriminator model to discriminate the new negative sample and the positive sample, and obtaining new probability, wherein in the discriminating process, the discriminator model can learn the distinction between the new negative sample and the positive sample, and the model parameters of the discriminator model are updated again according to the learned new distinction.
After calling the discriminator model to obtain new probability, taking the new probability as new rewards, combining with a reinforcement learning algorithm, calculating the value of the loss function according to new rewards, and then optimizing the model parameters of the intermediate model again according to the principle of reducing the calculated value of the loss function. Through repeated countermeasure learning between the discriminator model and the intermediate model, a balanced state between the discriminator model and the intermediate model can be caused, namely, corresponding answers corresponding to the questions determined by the intermediate model are infinitely approximate to standard answers, namely, the answers are infinitely close to human real language, the discriminator model cannot distinguish negative samples from positive samples, and the intermediate model in the balanced state can be used as a preset generation model.
Therefore, after the intermediate model is obtained, the intermediate model is not directly used as the preset generation model, but the idea of countermeasure learning is introduced, and the second training data set is adopted to perform optimization processing on the intermediate model, so that the preset generation model is obtained. The quality of the preset generation model can be further improved by optimizing the intermediate model, so that the generated answer determined by calling the preset generation model is closer to the human real language and has more authenticity.
S204, acquiring an initial question input by a user, and calling a preset retrieval model to determine a candidate answer set corresponding to the initial question from a preset knowledge base. The preset knowledge base comprises at least one question and one or more answers corresponding to each question, and the candidate answer set comprises at least one candidate answer.
The terminal can determine the initial question by acquiring voice information or text information, and then invokes a preset retrieval model to determine a candidate answer set corresponding to the initial question. Specifically, the specific implementation manner of calling the preset search model to determine the candidate answer set corresponding to the initial question from the preset knowledge base may include the following steps s11-s13:
s11, aiming at the initial problems, invoking a preset retrieval model to perform query processing in a preset knowledge base to determine at least one target problem, wherein at least one target word in the target problem is matched with at least one initial word in the initial problems.
Specifically, word and sentence splitting processing can be performed on the initial problem to obtain one or more initial words, and then word vector conversion is performed on each initial word to obtain word vectors of each initial word. Based on the word vector of each initial word, a preset retrieval model is called to find the matched word matched with each initial word in a preset knowledge base. For a target word, the target word includes an initial word and/or a matching word, and the target word is mapped to a target question containing the target word by using an inverted index method to determine the target question. The inverted index, which may also be referred to herein as an inverted index, a placement archive, or an inverted archive, is an indexing method used to store a mapping of the storage locations of a word in a document or group of documents under a full text search. By adopting the reverse index method, the target problem containing the target word can be quickly acquired according to the target word, and the retrieval rate is improved.
For example, the initial question is "which components of the computer exist", and the word and sentence splitting process is performed on the initial question, so that one or more initial words "computer" and "components" can be obtained. And then carrying out word vector conversion on each initial word, and calling a preset retrieval model to find a matching word matched with each initial word in a preset knowledge base based on the word vector of each initial word, for example, a tablet computer, a computer and a part. The initial word and/or the matching word are the target word, i.e., the target word may include one or more words of "computer," "component," "tablet," "computer," "part," etc. Taking the target word as an example of a computer, the target word may be mapped by adopting an inverted index method, and the mapping process refers to a process of searching for a target problem containing the target word. The inverted index method is adopted to find out the problem 'what the structure of the computer is', which contains the target word 'computer', so that the problem 'what the structure of the computer is' can be taken as a target problem; for another example, taking the target word as "computer" and "part" as an example, an inverted index method is used to find out which "parts of the problem" computer "contain the target word" computer "and" part ", so that which" parts of the problem "computer" can be used as the target problem.
And s12, respectively calculating the similarity between at least one target problem and the initial problem, and determining the target problem with the highest similarity.
Specifically, a similarity algorithm may be first used to calculate the similarity between at least one target problem and the initial problem, so as to obtain a plurality of similarities; the similarity algorithm herein may include, but is not limited to: BM25 algorithm, euclidean distance similarity algorithm, angle cosine similarity algorithm, pearson similarity algorithm, and so forth. After obtaining the plurality of similarities, the plurality of similarities may be ranked using a ranking function, where the ranking function may include, but is not limited to: rank Sort function, sort function, oracle Sort function, and so forth.
The ranking process may be a ranking process from high to low in similarity, or a ranking process from low to high in similarity. If the sorting process is a sorting process with high-to-low similarity, the target problem with the smallest sorting sequence number can be determined to be the target problem with highest similarity; if the sorting process is a sorting process with the similarity from low to high, the target problem with the largest sorting sequence number can be determined to be the target problem with the highest similarity.
And s13, acquiring at least one answer corresponding to the target question with the highest similarity from a preset knowledge base, and determining a candidate answer set according to the at least one answer.
After the target problem with the highest similarity is determined, at least one answer corresponding to the target problem with the highest similarity can be obtained from a preset knowledge base, and a candidate answer set is determined according to the at least one answer. In one embodiment, each answer corresponding to the target question with the highest acquired similarity may be used as a candidate answer, that is, a set formed by all answers corresponding to the target question with the highest acquired similarity is determined as a candidate answer set.
In still another embodiment, after at least one answer corresponding to the target question with the highest similarity is obtained from the preset knowledge base, the matching degree between each answer and the initial question may be calculated. And then selecting a preset number of answers to form a candidate answer set according to the sequence of the matching degree from high to low. The preset number of the answers can be determined according to the actual service requirement, for example, the preset number is 5, then the answers of the front top5 are all used as candidate answers according to the sequence of the matching degree from high to low, and the set formed by the candidate answers is determined as a candidate answer set.
For example, the answers corresponding to the target questions with highest similarity obtained from the preset knowledge base are 7 in total, and the 7 answers and the matching degree thereof are respectively: answer a (85% match), answer B (25% match), answer C (45% match), answer D (75% match), answer E (80% match), answer F (70% match) and answer G (60% match). The matching degree is from high to low as follows: answer A (degree of match 85%) > answer E (degree of match 80%) > answer D (degree of match 75%) > answer F (degree of match 70%) > answer G (degree of match 60%) > answer C (degree of match 45%) > answer B (degree of match 25%); and if the preset number is 5, the answers of the front top5 are all candidate answers, namely 'answer A', 'answer E', 'answer D', 'answer F', and 'answer G', namely 'answer A, answer E, answer D, answer F, and answer G'.
S205, a preset generating model is called to determine a generating answer set corresponding to the initial question, wherein the generating answer set comprises at least one generating answer, and the preset generating model is obtained by performing model training optimization by adopting a plurality of training data sets containing the question.
Specifically, a preset generation model can be called to determine at least one generated answer corresponding to the initial question, and the matching degree between each generated answer and the initial question is calculated according to a preset calculation rule; and then selecting a preset number of generated answers to form a generated answer set according to the sequence of the matching degree from high to low, wherein the preset number can be determined according to actual service requirements. For example, the preset number is 5, and then the generated answers of the front top5 are selected to form a generated answer set according to the sequence of the matching degree from high to low. In other embodiments, the preset generating model may be invoked to determine at least one generating answer corresponding to the initial question, and the set formed by all the generating answers determined by the preset generating model may be determined to be the generating answer set.
S206, calculating the candidate matching degree between each candidate answer in the candidate answer set and the initial question according to a preset calculation rule respectively to obtain at least one candidate matching degree, and solving a first average value of the at least one candidate matching degree.
S207, determining a target answer to be output from the candidate answer set or the generated answer set according to a preset determination rule and the first average value.
It should be noted that, the steps S206 to S207 in the embodiment of the present invention may refer to the steps S103 to S104 in the embodiment of the present invention, which is not described in detail.
In the embodiment of the invention, after the initial question input by the user is acquired, a preset retrieval model can be called to determine a candidate answer set corresponding to the initial question from a preset knowledge base, and a preset generation model is called to determine a generation answer set corresponding to the initial question. And then, calculating the candidate matching degree between each candidate answer in the candidate answer set and the initial question respectively to obtain at least one candidate matching degree, and solving a first average value of the at least one candidate matching degree. Finally, the target answer to be output can be determined from the candidate answer set or the generated answer set according to the first average value. The embodiment of the invention invokes the retrieval model and the generation model to determine the target answer, can avoid the long tail problem of the target answer and ensures the consistency and the rationality of the target answer. And determining whether to determine the target answer from the candidate answer set or the generated answer set according to the first average value can avoid the situation of wrong retrieval of the retrieval model, thereby improving accuracy.
Fig. 3a-3b are application scenario diagrams of an answer determining method based on data processing according to an embodiment of the present invention, where a user may open a user interface for performing a chat session with a terminal, as shown in fig. 3 a. The user may then enter the initial question at the user interface, as shown in fig. 3 b. After the terminal detects the input operation of the user, the terminal can acquire the initial question input by the user, call a preset retrieval model to determine a candidate answer set corresponding to the initial question from a preset knowledge base, and call a preset generation model to determine a generated answer set corresponding to the initial question. And then, respectively calculating the candidate matching degree between each candidate answer in the candidate answer set and the initial question according to a preset calculation rule to obtain at least one candidate matching degree, and solving a first average value of the at least one candidate matching degree. And finally, determining the target answer to be output from the candidate answer set or the generated answer set according to a preset determination rule and the first average value. After determining the target answer, the target answer may be output in a user interface to enable a chat session with the user, as shown in FIG. 3 b. And the retrieval model and the generation model are called to determine the target answer, so that the long-tail problem of the target answer can be avoided, and the consistency and the rationality of the target answer are ensured. And determining whether to determine the target answer from the candidate answer set or the generated answer set according to the first average value can avoid the situation of wrong retrieval of the retrieval model, thereby improving accuracy.
Fig. 4 is a schematic structural diagram of an answer determining apparatus based on data processing according to an embodiment of the present invention. As shown in fig. 4, the apparatus in the embodiment of the present invention may include:
the obtaining unit 101 is configured to obtain an initial question input by a user, and invoke a preset search model to determine a candidate answer set corresponding to the initial question from a preset knowledge base, where the preset knowledge base includes at least one question and one or more answers corresponding to each question, and the candidate answer set includes at least one candidate answer;
The obtaining unit 101 is configured to invoke a preset generating model to determine a generating answer set corresponding to the initial question, where the generating answer set includes at least one generating answer, and the preset generating model is obtained by performing model training optimization by using a plurality of training data sets including the question;
a calculating unit 102, configured to calculate candidate matching degrees between each candidate answer in the candidate answer set and the initial question according to a preset calculation rule, so as to obtain at least one candidate matching degree, and calculate a first average value of the at least one candidate matching degree;
and a determining unit 103, configured to determine, according to a preset determining rule, a target answer to be output from the candidate answer set or the generated answer set according to the first average value.
In one embodiment, the determining unit 103 is specifically configured to, when determining, according to a preset determination rule, a target answer to be output from the candidate answer set or the generated answer set according to the first average value:
judging whether the first average value is larger than a preset threshold value or not;
If the first average value is larger than the preset threshold value, selecting a candidate answer with highest candidate matching degree from the candidate answer set as a target answer;
And if the first average value is not greater than the preset threshold value, selecting a generated answer with highest generated matching degree from the generated answer set as a target answer, wherein the generated matching degree is the matching degree between the generated answer in the generated answer set and the initial question.
In still another embodiment, the determining unit 103 is specifically configured to, when determining, according to a preset determination rule, a target answer to be output from the candidate answer set or the generated answer set according to the first average value:
calculating the generation matching degree between each generation answer in the generation answer set and the initial question respectively to obtain at least one generation matching degree, and solving a second average value of the at least one generation matching degree;
If the first average value is larger than the second average value, selecting a candidate answer with highest candidate matching degree from the candidate answer set as a target answer;
If the first average value is smaller than the second average value, selecting a generated answer with highest generated matching degree from the generated answer set as a target answer;
And if the first average value is equal to the second average value, selecting a candidate answer with the highest candidate matching degree from the candidate answer set as a target answer, or selecting a generated answer with the highest generated matching degree from the generated answer set as the target answer.
In still another embodiment, when the obtaining unit 101 is configured to invoke a preset search model to determine, from a preset knowledge base, a candidate answer set corresponding to the initial question, the method is specifically configured to:
Aiming at the initial problems, invoking a preset retrieval model to perform query processing in the preset knowledge base so as to determine at least one target problem, wherein at least one target word in the target problem is matched with at least one initial word in the initial problems;
Respectively calculating the similarity between the at least one target problem and the initial problem, and determining the target problem with the highest similarity;
acquiring at least one answer corresponding to the target question with the highest similarity from the preset knowledge base, and determining a candidate answer set according to the at least one answer
In still another embodiment, when the obtaining unit 101 is configured to invoke a preset generation model to determine a generated answer set corresponding to the initial question, the obtaining unit is specifically configured to:
invoking a preset generation model to determine at least one generation answer corresponding to the initial question, and respectively calculating the matching degree between each generation answer and the initial question according to the preset calculation rule;
And selecting a preset number of generated answers to form a generated answer set according to the sequence of the matching degree from high to low.
In yet another embodiment, the obtaining unit 101 may further be configured to:
constructing a first training data set, wherein the first training data set comprises at least one pair of chat question-answer corpora, the at least one pair of chat question-answer corpora is obtained from at least one question-answer system, and each pair of chat question-answer corpora comprises a question and a corresponding standard answer;
Acquiring a pre-constructed initial model, and training the initial model by adopting the first training data set to obtain an intermediate model;
and acquiring a second training data set according to the first training data set, and adopting the second training data set to optimize the intermediate model to obtain a preset generation model.
In yet another embodiment, the obtaining unit 101 is specifically configured to, when configured to obtain the second training data set from the first training data set:
For each question in the first training data set, sequentially calling the intermediate model to determine a corresponding answer corresponding to each question, and taking the corresponding answers corresponding to all questions in the first training data set as negative samples;
Obtaining standard answers corresponding to each answer in the first training data set, and taking the standard answers corresponding to all questions in the first training data set as positive samples;
And taking the data set comprising the negative sample and the positive sample as a second training data set.
In the embodiment of the invention, after the initial question input by the user is acquired, a preset retrieval model can be called to determine a candidate answer set corresponding to the initial question from a preset knowledge base, and a preset generation model is called to determine a generation answer set corresponding to the initial question. And then, calculating the candidate matching degree between each candidate answer in the candidate answer set and the initial question respectively to obtain at least one candidate matching degree, and solving a first average value of the at least one candidate matching degree. Finally, the target answer to be output can be determined from the candidate answer set or the generated answer set according to the first average value. The embodiment of the invention invokes the retrieval model and the generation model to determine the target answer, can avoid the long tail problem of the target answer and ensures the consistency and the rationality of the target answer. And determining whether to determine the target answer from the candidate answer set or the generated answer set according to the first average value can avoid the situation of wrong retrieval of the retrieval model, thereby improving accuracy.
Based on the answer determining method and the device based on the data processing, the embodiment of the invention also provides a terminal which can be used for realizing the answer determining method based on the data processing. Fig. 5 is a schematic structural diagram of a terminal according to an embodiment of the present invention. As shown in fig. 5, the terminal includes an input device 201, an output device 202, a memory 203, and a processor 204, where the input device 201, the output device 202, the memory 203, and the processor 204 may be connected to each other, and the input device 201 may be used to obtain an initial problem transceiving message input by a user, and the input device 201 may correspond to the obtaining unit 101 in the above embodiment of the invention. The memory 203 may be used to store a computer program comprising program instructions, the computer program comprising program instructions. In yet another embodiment, the input device 201, the output device 202, the memory 203, and the processor 204 may be interconnected by way of a bus.
Those skilled in the art will appreciate that implementing all or part of the above described embodiment methods may be accomplished by computer programs in hardware associated with the instructions, the programs being stored on a computer readable storage medium, the programs comprising at least one program instruction loaded by the processor 204 and adapted to perform the steps of:
Acquiring an initial question input by a user, and calling a preset retrieval model to determine a candidate answer set corresponding to the initial question from a preset knowledge base, wherein the preset knowledge base comprises at least one question and one or more answers corresponding to each question, and the candidate answer set comprises at least one candidate answer;
Invoking a preset generation model to determine a generation answer set corresponding to the initial question, wherein the generation answer set comprises at least one generation answer, and the preset generation model is obtained by performing model training optimization by adopting a plurality of training data sets containing the question;
Calculating candidate matching degrees between each candidate answer in the candidate answer set and the initial question according to a preset calculation rule to obtain at least one candidate matching degree, and solving a first average value of the at least one candidate matching degree;
and determining a target answer to be output from the candidate answer set or the generated answer set according to a preset determination rule and the first average value.
In one embodiment, the at least one program instruction may be loaded by the processor 204 and configured to perform:
judging whether the first average value is larger than a preset threshold value or not;
If the first average value is larger than the preset threshold value, selecting a candidate answer with highest candidate matching degree from the candidate answer set as a target answer;
And if the first average value is not greater than the preset threshold value, selecting a generated answer with highest generated matching degree from the generated answer set as a target answer, wherein the generated matching degree is the matching degree between the generated answer in the generated answer set and the initial question.
In yet another embodiment, the at least one program instruction may be loaded by the processor 204 and configured to perform:
calculating the generation matching degree between each generation answer in the generation answer set and the initial question respectively to obtain at least one generation matching degree, and solving a second average value of the at least one generation matching degree;
If the first average value is larger than the second average value, selecting a candidate answer with highest candidate matching degree from the candidate answer set as a target answer;
If the first average value is smaller than the second average value, selecting a generated answer with highest generated matching degree from the generated answer set as a target answer;
And if the first average value is equal to the second average value, selecting a candidate answer with the highest candidate matching degree from the candidate answer set as a target answer, or selecting a generated answer with the highest generated matching degree from the generated answer set as the target answer.
In yet another embodiment, the at least one program instruction may be loaded by the processor 204 and configured to perform, when the set of candidate answers corresponding to the initial question is determined from a preset knowledge base by invoking a preset retrieval model:
Aiming at the initial problems, invoking a preset retrieval model to perform query processing in the preset knowledge base so as to determine at least one target problem, wherein at least one target word in the target problem is matched with at least one initial word in the initial problems;
Respectively calculating the similarity between the at least one target problem and the initial problem, and determining the target problem with the highest similarity;
And acquiring at least one answer corresponding to the target question with the highest similarity from the preset knowledge base, and determining a candidate answer set according to the at least one answer.
In yet another embodiment, when the predetermined generation model is invoked to determine the generated answer set corresponding to the initial question, the at least one program instruction may be loaded by the processor 204 and used to execute:
Invoking a preset generation model to determine at least one generation answer corresponding to the initial question, and respectively calculating the matching degree between each generation answer and the initial question according to a preset calculation rule;
And selecting a preset number of generated answers to form a generated answer set according to the sequence of the matching degree from high to low.
In yet another embodiment, the at least one program instruction is further loadable by the processor 204 and configured to perform:
constructing a first training data set, wherein the first training data set comprises at least one pair of chat question-answer corpora, the at least one pair of chat question-answer corpora is obtained from at least one question-answer system, and each pair of chat question-answer corpora comprises a question and a corresponding standard answer;
Acquiring a pre-constructed initial model, and training the initial model by adopting the first training data set to obtain an intermediate model;
and acquiring a second training data set according to the first training data set, and adopting the second training data set to optimize the intermediate model to obtain a preset generation model.
In yet another embodiment, the at least one program instruction is loadable by the processor and configured to perform, when the second training data set is obtained from the first training data set:
For each question in the first training data set, sequentially calling the intermediate model to determine a corresponding answer corresponding to each question, and taking the corresponding answers corresponding to all questions in the first training data set as negative samples;
Obtaining standard answers corresponding to each answer in the first training data set, and taking the standard answers corresponding to all questions in the first training data set as positive samples;
And taking the data set comprising the negative sample and the positive sample as a second training data set.
In the embodiment of the invention, after the initial question input by the user is acquired, a preset retrieval model can be called to determine a candidate answer set corresponding to the initial question from a preset knowledge base, and a preset generation model is called to determine a generation answer set corresponding to the initial question. And then, calculating the candidate matching degree between each candidate answer in the candidate answer set and the initial question respectively to obtain at least one candidate matching degree, and solving a first average value of the at least one candidate matching degree. Finally, the target answer to be output can be determined from the candidate answer set or the generated answer set according to the first average value. The embodiment of the invention invokes the retrieval model and the generation model to determine the target answer, can avoid the long tail problem of the target answer and ensures the consistency and the rationality of the target answer. And determining whether to determine the target answer from the candidate answer set or the generated answer set according to the first average value can avoid the situation of wrong retrieval of the retrieval model, thereby improving accuracy.
The embodiment of the invention also provides a computer storage medium which stores a computer program. The computer program comprises at least one program instruction loadable by a processor and adapted to perform the answer determination method based on data processing as described above.
The computer storage medium is a memory device for storing programs and data. It is to be understood that the computer storage media herein may include built-in storage media in the server, or may include extended storage media supported by the server. In one embodiment, the computer storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), or a random access Memory (Random Access Memory, RAM), or the like.
The above disclosure is only a few examples of the present application, and it is not intended to limit the scope of the claims, and those skilled in the art will understand that all or a portion of the above embodiments may be implemented and equivalents may be modified according to the claims of the present application.
Claims (8)
1. A method for answer determination based on data processing, comprising:
Acquiring an initial question input by a user, and calling a preset retrieval model to determine a candidate answer set corresponding to the initial question from a preset knowledge base, wherein the preset knowledge base comprises at least one question and one or more answers corresponding to each question, and the candidate answer set comprises at least one candidate answer;
Invoking a preset generation model to determine a generation answer set corresponding to the initial question, wherein the generation answer set comprises at least one generation answer, and the preset generation model is obtained by performing model training optimization by adopting a plurality of training data sets containing the question;
Calculating candidate matching degrees between each candidate answer in the candidate answer set and the initial question according to a preset calculation rule to obtain at least one candidate matching degree, and solving a first average value of the at least one candidate matching degree;
judging whether the first average value is larger than a preset threshold value or not; if the first average value is larger than the preset threshold value, selecting a candidate answer with highest candidate matching degree from the candidate answer set as a target answer; if the first average value is not greater than the preset threshold value, selecting a generated answer with highest generated matching degree from the generated answer set as a target answer, wherein the generated matching degree is the matching degree between the generated answer in the generated answer set and the initial question; or alternatively
Calculating the generation matching degree between each generation answer in the generation answer set and the initial question respectively to obtain at least one generation matching degree, and solving a second average value of the at least one generation matching degree; if the first average value is larger than the second average value, selecting a candidate answer with highest candidate matching degree from the candidate answer set as a target answer; if the first average value is smaller than the second average value, selecting a generated answer with highest generated matching degree from the generated answer set as a target answer; and if the first average value is equal to the second average value, selecting a candidate answer with the highest candidate matching degree from the candidate answer set as a target answer, or selecting a generated answer with the highest generated matching degree from the generated answer set as the target answer.
2. The method of claim 1, wherein the invoking the preset search model to determine the candidate answer set corresponding to the initial question from a preset knowledge base comprises:
Aiming at the initial problems, invoking a preset retrieval model to perform query processing in the preset knowledge base so as to determine at least one target problem, wherein at least one target word in the target problem is matched with at least one initial word in the initial problems;
Respectively calculating the similarity between the at least one target problem and the initial problem, and determining the target problem with the highest similarity;
And acquiring at least one answer corresponding to the target question with the highest similarity from the preset knowledge base, and determining a candidate answer set according to the at least one answer.
3. The method of claim 1, wherein the invoking the preset generation model to determine the generated answer set corresponding to the initial question comprises:
invoking a preset generation model to determine at least one generation answer corresponding to the initial question, and respectively calculating the matching degree between each generation answer and the initial question according to the preset calculation rule;
And selecting a preset number of generated answers to form a generated answer set according to the sequence of the matching degree from high to low.
4. The method of claim 1, wherein the method further comprises:
constructing a first training data set, wherein the first training data set comprises at least one pair of chat question-answer corpora, the at least one pair of chat question-answer corpora is obtained from at least one question-answer system, and each pair of chat question-answer corpora comprises a question and a corresponding standard answer;
Acquiring a pre-constructed initial model, and training the initial model by adopting the first training data set to obtain an intermediate model;
and acquiring a second training data set according to the first training data set, and adopting the second training data set to optimize the intermediate model to obtain a preset generation model.
5. The method of claim 4, wherein the acquiring a second training data set from the first training data set comprises:
For each question in the first training data set, sequentially calling the intermediate model to determine a corresponding answer corresponding to each question, and taking the corresponding answers corresponding to all questions in the first training data set as negative samples;
Obtaining standard answers corresponding to each answer in the first training data set, and taking the standard answers corresponding to all questions in the first training data set as positive samples;
And taking the data set comprising the negative sample and the positive sample as a second training data set.
6. An answer determining apparatus based on data processing, comprising:
The system comprises an acquisition unit, a search unit and a search unit, wherein the acquisition unit is used for acquiring an initial question input by a user, and invoking a preset search model to determine a candidate answer set corresponding to the initial question from a preset knowledge base, the preset knowledge base comprises at least one question and one or more answers corresponding to each question, and the candidate answer set comprises at least one candidate answer;
The acquisition unit is used for calling a preset generation model to determine a generation answer set corresponding to the initial question, wherein the generation answer set comprises at least one generation answer, and the preset generation model is obtained by performing model training optimization by adopting a plurality of training data sets containing the question;
the computing unit is used for respectively computing candidate matching degrees between each candidate answer in the candidate answer set and the initial question according to a preset computing rule so as to obtain at least one candidate matching degree, and solving a first average value of the at least one candidate matching degree;
The determining unit is used for determining a target answer to be output from the candidate answer set or the generated answer set according to a preset determining rule and the first average value;
Wherein, the determining unit is specifically configured to:
judging whether the first average value is larger than a preset threshold value or not; if the first average value is larger than the preset threshold value, selecting a candidate answer with highest candidate matching degree from the candidate answer set as a target answer; if the first average value is not greater than the preset threshold value, selecting a generated answer with highest generated matching degree from the generated answer set as a target answer, wherein the generated matching degree is the matching degree between the generated answer in the generated answer set and the initial question; or alternatively
Calculating the generation matching degree between each generation answer in the generation answer set and the initial question respectively to obtain at least one generation matching degree, and solving a second average value of the at least one generation matching degree; if the first average value is larger than the second average value, selecting a candidate answer with highest candidate matching degree from the candidate answer set as a target answer; if the first average value is smaller than the second average value, selecting a generated answer with highest generated matching degree from the generated answer set as a target answer; and if the first average value is equal to the second average value, selecting a candidate answer with the highest candidate matching degree from the candidate answer set as a target answer, or selecting a generated answer with the highest generated matching degree from the generated answer set as the target answer.
7. A terminal comprising an input device, an output device, a memory, and a processor, the input device, the output device, and the memory being interconnected, wherein the memory is configured to store a computer program comprising program instructions, the processor being configured to invoke the program instructions to perform the method of any of claims 1-5.
8. A computer readable storage medium, characterized in that the computer storage medium stores a computer program comprising program instructions which, when executed by a processor, cause the processor to perform the method of any of claims 1-5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811364713.0A CN109492085B (en) | 2018-11-15 | 2018-11-15 | Answer determination method, device, terminal and storage medium based on data processing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811364713.0A CN109492085B (en) | 2018-11-15 | 2018-11-15 | Answer determination method, device, terminal and storage medium based on data processing |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109492085A CN109492085A (en) | 2019-03-19 |
| CN109492085B true CN109492085B (en) | 2024-05-14 |
Family
ID=65695856
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811364713.0A Active CN109492085B (en) | 2018-11-15 | 2018-11-15 | Answer determination method, device, terminal and storage medium based on data processing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109492085B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110414246B (en) * | 2019-06-19 | 2023-05-30 | 平安科技（深圳）有限公司 | Shared file security management method, device, terminal and storage medium |
| CN110489730A (en) * | 2019-08-14 | 2019-11-22 | 腾讯科技（深圳）有限公司 | Text handling method, device, terminal and storage medium |
| CN110543552B (en) * | 2019-09-06 | 2022-06-07 | 网易（杭州）网络有限公司 | Conversation interaction method and device and electronic equipment |
| CN111339275B (en) * | 2020-02-27 | 2023-05-12 | 深圳大学 | Answer information matching method, device, server and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106682387A (en) * | 2016-10-26 | 2017-05-17 | 百度国际科技（深圳）有限公司 | Method and device used for outputting information |
| CN107133303A (en) * | 2017-04-28 | 2017-09-05 | 百度在线网络技术（北京）有限公司 | Method and apparatus for output information |
| CN108491433A (en) * | 2018-02-09 | 2018-09-04 | 平安科技（深圳）有限公司 | Chat answer method, electronic device and storage medium |
| CN108509463A (en) * | 2017-02-28 | 2018-09-07 | 华为技术有限公司 | A kind of answer method and device of problem |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104598445B (en) * | 2013-11-01 | 2019-05-10 | 腾讯科技（深圳）有限公司 | Automatically request-answering system and method |
-
2018
- 2018-11-15 CN CN201811364713.0A patent/CN109492085B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106682387A (en) * | 2016-10-26 | 2017-05-17 | 百度国际科技（深圳）有限公司 | Method and device used for outputting information |
| CN108509463A (en) * | 2017-02-28 | 2018-09-07 | 华为技术有限公司 | A kind of answer method and device of problem |
| CN107133303A (en) * | 2017-04-28 | 2017-09-05 | 百度在线网络技术（北京）有限公司 | Method and apparatus for output information |
| CN108491433A (en) * | 2018-02-09 | 2018-09-04 | 平安科技（深圳）有限公司 | Chat answer method, electronic device and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109492085A (en) | 2019-03-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109492085B (en) | Answer determination method, device, terminal and storage medium based on data processing | |
| CN107480143B (en) | Method and system for segmenting conversation topics based on context correlation | |
| KR20210038449A (en) | Question and answer processing, language model training method, device, equipment and storage medium | |
| CN110148416A (en) | Audio recognition method, device, equipment and storage medium | |
| WO2017181834A1 (en) | Intelligent question and answer method and device | |
| US20190205743A1 (en) | System and method for detangling of interleaved conversations in communication platforms | |
| EP3575988A1 (en) | Method and device for retelling text, server, and storage medium | |
| CN109325108B (en) | Query processing method, device, server and storage medium | |
| US11853689B1 (en) | Computer-implemented presentation of synonyms based on syntactic dependency | |
| US9811517B2 (en) | Method and system of adding punctuation and establishing language model using a punctuation weighting applied to chinese speech recognized text | |
| CN110134777B (en) | Question duplication eliminating method and device, electronic equipment and computer readable storage medium | |
| US11120214B2 (en) | Corpus generating method and apparatus, and human-machine interaction processing method and apparatus | |
| CN114757176A (en) | Method for obtaining target intention recognition model and intention recognition method | |
| CN112686051B (en) | Semantic recognition model training method, recognition method, electronic device and storage medium | |
| US20230065965A1 (en) | Text processing method and apparatus | |
| CN110717021A (en) | Input text and related device for obtaining artificial intelligence interview | |
| CN114003682A (en) | Text classification method, device, equipment and storage medium | |
| CN108268443B (en) | Method and device for determining topic point transfer and acquiring reply text | |
| CN114116997A (en) | Knowledge question answering method, knowledge question answering device, electronic equipment and storage medium | |
| CN117421403A (en) | Intelligent dialogue method and device and electronic equipment | |
| CN109684357B (en) | Information processing method and device, storage medium and terminal | |
| US20220245340A1 (en) | Electronic device for processing user's inquiry, and operation method of the electronic device | |
| CN115906797A (en) | Text entity alignment method, device, equipment and medium | |
| US11526663B2 (en) | Methods, apparatuses, devices, and computer-readable storage media for determining category of entity | |
| CN112905752A (en) | Intelligent interaction method, device, equipment and storage 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |