CN110502621A - Answering method, question and answer system, computer equipment and storage medium - Google Patents
Answering method, question and answer system, computer equipment and storage medium Download PDFInfo
- Publication number
- CN110502621A CN110502621A CN201910593110.6A CN201910593110A CN110502621A CN 110502621 A CN110502621 A CN 110502621A CN 201910593110 A CN201910593110 A CN 201910593110A CN 110502621 A CN110502621 A CN 110502621A
- Authority
- CN
- China
- Prior art keywords
- entity
- candidate
- triple
- input information
- name
- 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.)
- Granted
Links
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
- G06F16/3347—Query execution using vector based model
-
- 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
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Human Computer Interaction (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses answering method, device, computer equipment and storage mediums, comprising: obtains the input information of user;Name entity in identification input information, and entity link candidate entity corresponding with name entity into Chinese knowledge mapping will be named, entity pair is formed, wherein entity is to including naming entity and candidate entity;The candidate relationship of candidate entity is matched in Chinese knowledge mapping by relational model;According to entity to and candidate relationship, candidate triple is formed, wherein candidate triple includes name entity, candidate entity and candidate relationship;Based on study order models, the corresponding ranking results of each candidate triple are obtained;And according to ranking results, the Chinese knowledge mapping is inquired, to obtain the answer of input information.This method can efficiently use external resource, can provide a large amount of language ambience information by text mining, and based on study order models, preferable answer can also be obtained when question and answer corpus data is less.
Description
Technical field
The present invention relates to field of computer technology more particularly to a kind of answering method, question and answer system, computer equipment and deposit
Storage media.
Background technique
Question answering system is a kind of advanced form of information retrieval system, it can be answered with accurate, succinct natural language and be used
The problem of family is proposed with natural language.Traditional question answering system is divided into question sentence processing and two large divisions is retrieved in answer.Wherein, question sentence
The basis of processing is participle.Answer retrieval mostly uses scoring, i.e., a series of candidate answers are chosen from mass text data,
Then building selection function chooses immediate answer from candidate answers.And this traditional question and answer system in processing because grow
Text noun and building selection function difference and there is different degrees of mistake.
In this case, the question answering system of knowledge based map is come into being.Currently, the question and answer system of knowledge based map
Main research direction of uniting has three classes.The first kind: regular pattern composite, this type determine user's question sentence by fixed rule
It whether is some fact in inquiry knowledge base.Second class: Template Learning, this type collects a large amount of templates, by
The good corresponding knowledge base of mark is true, and learns the probability that a natural language question sentence corresponds to some template by mass data.
Third class: the semantic matches based on deep learning learn some pass in a question sentence and knowledge mapping by neural network model
The semantic similarity of system, wherein question sentence had done Entity recognition and had replaced entity in question sentence with additional character.
The knowledge base question answering system accuracy of regular pattern composite is very high, but and it is not flexible, it is every one kind problem require to write one
Rule, and Template Learning and deep learning method generally require to be learnt according to extensive question and answer corpus, it is difficult to it is first in exploitation
Phase is just applied in the vertical field of a question and answer data scarcity.
Summary of the invention
In view of this, the present invention proposes a kind of answering method, question and answer system, computer equipment and storage medium, Neng Gou
In the case that question and answer corpus data is less, obtains one and accurately answer.
Firstly, to achieve the above object, the present invention proposes a kind of answering method, the answering method comprising steps of
Obtain the input information of user;
Identify the name entity in the input information, and by the name entity link into the Chinese knowledge mapping
Candidate's entity corresponding with the name entity, forms entity pair, wherein the entity is to including the name entity and institute
State candidate entity;
The candidate relationship of the candidate entity is matched in the Chinese knowledge mapping by relational model;
According to the entity to the candidate relationship, form candidate triple;Wherein the candidate triple includes institute
State name entity, the candidate entity and the candidate relationship;
The corresponding ranking results of each candidate's triple are obtained based on study order models;And
According to the ranking results inquiry Chinese knowledge mapping, to obtain the answer of the input information.
Further, the name entity in the identification input information specifically includes:
The input information is labeled, annotation results are obtained;And
According to the annotation results, the name entity in the input information is identified by Recognition with Recurrent Neural Network model.
Further, the name entity in the identification input information, and by the name entity link to described
Candidate's entity corresponding with the name entity, the step for forming entity pair include: in Chinese knowledge mapping
The similarity between name entity described in each entity pair and the candidate entity is calculated, wherein described similar
Degree is obtained according to chinese character similarity, pinyin character similarity, term vector similarity and the concerned degree of entity;
It is arranged the names with obtaining each entity to corresponding to each entity to sequence according to each similarity;And
The corresponding entity pair is chosen according to described arrange the names.
Further, the relationship templates include between first instance, second instance and first instance and second instance
Relationship.
Further, described based on study order models, obtain the corresponding ranking results tool of each candidate's triple
Body includes:
Calculate the corresponding each feature vector of each triple;And
Each described eigenvector is input in the study order models corresponding to obtain each candidate triple
Ranking results.
Further, the step of each feature vector for calculating each triple includes:
The first similarity feature between the name entity and the candidate entity is calculated according to the triple;
The name entity in the input information is removed to obtain remaining word, and calculates the remaining word and same
The second similarity feature between adopted word and context vocabulary;
According to the input information generate high dimension vector, wherein the high dimension vector be according in the input information whether
It is generated there are default vocabulary;And
According to the first similarity feature, the second similarity feature and the high dimension vector, the feature is generated
Vector.
Further, the study order models are formed by by training first sample and each candidate triple
What the second sample obtained, wherein first sample is the triple being made of the model answer of the input information.
A kind of question and answer system of the Chinese knowledge mapping based on study sequence, the question and answer system include:
First obtains module, for obtaining the input information of user;
Identification and link module, the name entity in the input information for identification, and by the name entity link
Into the Chinese knowledge mapping, candidate entity corresponding with the name entity, forms entity pair, wherein the entity pair
Including the name entity and the candidate entity;
Matching module, for matching the candidate of the candidate entity in the Chinese knowledge mapping by relational model
Relationship;
Form module, for according to the entity to the candidate relationship, form candidate triple;The wherein candidate
Triple includes the name entity, the candidate entity and the candidate relationship;
Second acquisition module, for obtaining the corresponding sequence knot of each candidate's triple based on study order models
Fruit;And
Third obtains module, for inquiring the Chinese knowledge mapping according to the ranking results, to obtain the input
The answer of information.
To achieve the above object, it the present invention also provides a kind of computer equipment, including memory, processor and is stored in
On memory and the computer program that can run on the processor, the processor are realized when executing the computer program
The step of above method.
To achieve the above object, the present invention also provides computer readable storage mediums, are stored thereon with computer program, institute
State the step of above method is realized when computer program is executed by processor.
Compared to traditional technology, answering method, computer equipment and the storage of knowledge based map proposed by the invention
Medium can effectively utilize external resource, and the synonym or online text of the relationship fact are efficiently used by width learning model
The external resources such as word, this portion of external resource quick by text mining or directly in the way of Chinese word body etc. can obtain
It arrives.Also by the combination of width learning model and deep learning model, data volume needed for can reduce model, in training data
It also can be preferably exported when less as a result, this has when the knowledge mapping question and answer in exploitation new vertical field
Very important meaning.
Detailed description of the invention
Fig. 1 is the flow diagram of the answering method of first embodiment of the invention;
Fig. 2 is the flow diagram of the answering method of second embodiment of the invention;
Fig. 3 is the flow diagram of the answering method of third embodiment of the invention;
Fig. 4 is the flow diagram of the answering method of fourth embodiment of the invention;
Fig. 5 is the flow diagram of the answering method of fifth embodiment of the invention;
Fig. 6 is the block diagram of the question and answer system of sixth embodiment of the invention;
Fig. 7 is the block diagram of the question and answer system of seventh embodiment of the invention;And
Fig. 8 is the block diagram of synonym collector unit in the question and answer system of eighth embodiment of the invention.
The embodiments will be further described with reference to the accompanying drawings for the realization, the function and the advantages of the object of the present invention.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that described herein, specific examples are only used to explain the present invention, not
For limiting the present invention.Based on the embodiments of the present invention, those of ordinary skill in the art are not before making creative work
Every other embodiment obtained is put, shall fall within the protection scope of the present invention.
It should be noted that the description for being related to " first ", " second " etc. in the present invention is used for description purposes only, and cannot
It is interpreted as its relative importance of indication or suggestion or implicitly indicates the quantity of indicated technical characteristic.Define as a result, " the
One ", the feature of " second " can explicitly or implicitly include at least one of the features.In addition, the skill between each embodiment
Art scheme can be combined with each other, but must be based on can be realized by those of ordinary skill in the art, when technical solution
Will be understood that the combination of this technical solution is not present in conjunction with there is conflicting or cannot achieve when, also not the present invention claims
Protection scope within.
Referring to FIG. 1, providing a kind of answering method in first embodiment.The answering method includes:
Step S110: the input information of user is obtained.
Wherein, input information can be natural language retrieval sentence (such as question sentence), as user inputs question sentence on search website:
" what medicine cough needs to eat " the present embodiment inputs the mode of information without limitation to acquisition.
Step S120: the name entity in the identification input information, and by the name entity link to the Chinese
Candidate entity in knowledge mapping forms entity pair, wherein the entity is to including the name entity and the candidate entity.
Specifically, by carrying out sequence labelling, then root to input information using mark set method, Recognition with Recurrent Neural Network model
The identification of name entity is completed according to the result of sequence labelling (specific steps will be discussed in detail in a second embodiment).Such as
" what medicine cough needs to eat ", first pass through BIO mark set method and it be labeled, obtain the question sentence according to annotation results
Vector information, then using the vector information as the input of Recognition with Recurrent Neural Network model, to identify " cough " this name reality
Body.Then, which is corresponded to globally unique identifier (the Globally Unique in Chinese knowledge mapping
Identifier, GUID), thus corresponding candidate's entity, such as cough into knowledge mapping by the name entity link.In addition,
Each of knowledge mapping candidate's entity uniquely corresponds to a GUID, can be distinguished in Chinese knowledge mapping by the GUID
Different candidate's entities.
Wherein, Chinese knowledge mapping is a kind of new technique for storing labyrinth information.It is deposited in Chinese knowledge mapping
A host of facts type knowledge is stored up, relation information of the storage inside between entity and entity.Chinese knowledge mapping is mostly with RDF
The format memory data of (Resource Description Framework), a fact are represented as (S, P, an O) three
Tuple, shaped like (subject, predicate, object), wherein S and O is expressed as entity, and O is sometimes also indicated as attribute value, P
Indicate the relationship between S and O.Entity link be solve name entity ambiguity problem a kind of important method, this method pass through by
Have ambiguous entity denotion item and is linked to the elimination for realizing entity ambiguity in given knowledge mapping.
In addition, because there is alias or other information in name entity, each of Chinese knowledge mapping candidate's entity and
Its corresponding name and alias, obtain Alias information, and the reversed dictionary for constructing alias to candidate entity is used for entity link.Building
When dictionary, needs to carry out alias character string unitized processing, for example be converted to lowercase character, leave out spcial character etc., and
The entity in alias dictionary is ranked up as popularity in the frequency that knowledge mapping kind occurs by obtaining entity.It identifies
After naming Entity recognition, we are searched in alias dictionary using name entity and obtain candidate entity, and knowing according to entity
Name degree chooses conduct candidate's entity in the top.
Step S130: the candidate of candidate entity is matched in the Chinese knowledge mapping by relationship templates and is closed
System.
Specifically, relationship templates are understood expressed by the input information (such as question sentence) of user by natural language understanding technology
Semanteme, and matched with the relationship P in the triple (S, P, O) in Chinese knowledge mapping, which is determined with this
Corresponding candidate relationship in the semantic and Chinese knowledge mapping of expression.Wherein, relationship templates include first instance, second instance and
Relationship between first instance and second instance.Relationship templates be by extracting some triples in Chinese knowledge mapping, and
Relation information is extracted from these triples, to obtain according to the process training of these relation informations corresponding with these relation informations
Relationship templates.
Step S140: according to the entity to the candidate relationship, form candidate triple;The wherein candidate ternary
Group includes the name entity, the candidate entity and the candidate relationship.
Specifically, the name entity identified using above-mentioned steps, the name entity are corresponding in Chinese knowledge mapping
Candidate entity and candidate relationship, to form each candidate triple.
Step S150: the corresponding ranking results of each candidate's triple are obtained based on study order models.
Specifically, each candidate triple is converted into corresponding vector information, and the input as study order models, warp
The series of computation of the study order models is crossed, thus output ranking results corresponding with each candidate's triple.Wherein sort
As a result can be more more forward more accurate according to ranking, more more inaccurate rule is arranged ranking rearward, it is also possible to other modes,
The present embodiment does not limit.
Wherein, study order models are calculated using study sort algorithm.Learn sort algorithm (Learning
Torank, LTR) be a kind of supervised learning (SupervisedLearning, SL) sort method.LTR generally has three classes
Method: single document method (Pointwise), document is to method (Pairwise), lists of documents method (Listwise).This implementation
Study sort algorithm is using document to method (Pairwise) in example.
In one embodiment, study order models are to be formed by second by training first sample and each candidate triple
What sample obtained, wherein first sample is the triple being made of the model answer of inputted information.For example, according to a question sentence
Model answer triple (name entity, candidate entity, candidate relationship), take 10 times at random in Chinese knowledge mapping
Entity is selected, is acquiring candidate relationship according to this candidate entity, finally obtaining 50 triples, (name entity, is waited at candidate entity
Select relationship) composition negative sample (N).Wherein the triple (name entity, candidate entity, candidate relationship) of the model answer is positive
Sample (P).By a combination in the positive sample (P) and negative sample (N), generate two samples, be (P, N) sample and (N,
P) sample.The label of (P, N) sample is 1, and the label of (N, P) sample is 0.Learn order models can according to the sample training and
It arrives.
Step S160: according to the ranking results inquiry Chinese knowledge mapping, to obtain answering for the input information
Case.
Specifically, according to the ranking results of each candidate triple, candidate triple of the ranking before preset value is chosen, then
These selected candidate triples are converted into the query language in Chinese knowledge mapping, to hold in Chinese knowledge mapping
The row query statement, returns to the corresponding answer of input information after inquiry.
In short, the answering method, which can use study order models, effectively utilizes external resource, in question and answer corpus data
In the case where less, for user the problem of can also obtain accurate answer.
In a second embodiment, referring to FIG. 2, being identified in the input information in step S120 in first embodiment
Name entity the step for include:
Step S210 is labeled the input information, obtains annotation results.
Assuming that the question sentence of user's input is q:X=(x1,x2,…,xn),xiIt indicates each word in question sentence, is marked using BIEO
Injecting method marks each word in question sentence, and " B " is the beginning for naming entity, and " I " indicates inside name entity that " E " indicates name
The end of entity, " O " indicate it is not name entity.Y=(y1,y2,…,yn) indicate annotation results, it is obtained by the mask method
Annotation results score are as follows:
Wherein matrix P ∈ RK×nThe as state eigenmatrix of condition random field, PI, jIndicate that j-th of word in sentence is marked
Note is the score of i-th kind of label, A ∈ R(K+2)×(k+2)Indicate state-transition matrix, elements AI, jIt indicates to shift from i-th kind of label
To the score of jth kind label.Mask method is also possible to other mask methods, such as BIO, BIOES etc., and the present embodiment does not limit
It is fixed.For example, it is as follows to the mark situation of question sentence to mark set method by BIEO: fishing (O) fish (O) than (O) match (O) in (O) tall building (B-
LOC) (O) row (O) is lifted in the city door (I-LOC) (E-LOC).Using mark collection be in order to reduce noise as far as possible, thus identification and
The entity accuracy rate extracted is higher.
Step S220 is identified in the input information according to the annotation results by Recognition with Recurrent Neural Network model
Name entity.
Specifically, by the annotation results in above-mentioned steps, so that the annotation results of each word are got, further according to the mark
As a result, obtaining the vector information of each word.For example, the annotation results of each word are converted to one-hot vector, then by each word
One-hot DUAL PROBLEMS OF VECTOR MAPPING be the dense word vector of low-dimensional, the word vector of each word in the sentence is then subjected to successively group
Arrangement is closed, to obtain the vector information of entire sentence.The vector information of entire sentence is input to Recognition with Recurrent Neural Network mould again
In type, that is, it may recognize that the name entity in question sentence.Wherein Recognition with Recurrent Neural Network model can calculate the input information in question sentence
In the corresponding label of each word probability and obtain optimal sequence label.Wherein the optimal sequence label is the life identified
Name entity.Recognition with Recurrent Neural Network model can be two-way long short-term memory Recognition with Recurrent Neural Network model, be also possible to condition random
Field model etc., the present embodiment to it without limitation.
In third embodiment, referring to FIG. 3, the answering method is also in one embodiment after step S130
Include:
Step S310 calculates the similarity between name entity described in each entity pair and the candidate entity,
Described in similarity be according to chinese character similarity, pinyin character similarity, term vector similarity and entity is concerned spends
It arrives.
Specifically, chinese character similarity, the pinyin character between entity centering name entity and candidate entity are calculated
Similarity, term vector similarity and the concerned degree of entity, comprehensive each similarity is to obtain each entity to corresponding similarity.
Wherein, similarity it is higher illustrate to name it is more similar between entity and candidate entity.The method for calculating similarity has based on bag of words mould
Type, after it will name entity and candidate entity vectorization, it is converted into the distance calculated in space, apart from smaller similarity
It is higher;Also have and press from both sides cosine of an angle between calculating two vectors.The size of the cosine can directly reflect similarity, i.e. cosine is smaller similar
It spends higher;The present embodiment to calculate similarity method without limitation.The present embodiment passes through in chinese character, pinyin character, word
The different aspects such as vector and attention rate calculate separately similarity, and finally synthesis obtains similarity again, so as to more accurately sentence
Similarity degree between disconnected name entity and candidate entity, is also beneficial to find optimal candidate entity.
Step S320, according to each similarity to each entity to sequence, to obtain each entity to corresponding arrangement
Ranking.
Specifically, the similarity calculated according to above-mentioned steps, thus according to similarity size to each entity into
Row sequence, and then each entity is obtained to arranging the names in all entities pair.Wherein, similarity is higher, illustrates candidate entity
Higher with the matching degree of name entity, similarity is lower to be illustrated candidate entity and names the matching degree of entity lower.
Step S330 chooses the corresponding entity pair according to described arrange the names.
Specifically, each entity pair to arrange the names before default ranking is chosen.Wherein, default ranking can be according to practical feelings
Condition is set.In the present embodiment, presetting ranking is the 10th, selected so as to select before ranking ten entity pair
The candidate entity of the entity centering taken is also closer to the name entity in input information.
In the 4th embodiment, referring to FIG. 4, step S150 includes in one embodiment;
Step S410 calculates the corresponding each feature vector of each triple.
Specifically, the name entity in each triple, candidate entity and candidate relationship are converted into one-hot vector, then
It is mapped as the dense word vector of low-dimensional, finally each word vector is arranged again, gets the feature vector of each triple.
Each described eigenvector is input in the study order models to obtain each candidate ternary by step S420
The corresponding ranking results of group.
Specifically, using each feature vector as the input of study order models, the calculating through overfitting order models, output
The corresponding ranking results of each triple.
In the 5th embodiment, referring to FIG. 5, step S410 includes: in the 4th implementation
Step S510 calculates the first similarity between the name entity and the candidate entity according to the triple
Feature.
Specifically, for triple (name entity, candidate entity, candidate relationship), name entity and candidate entity are calculated
Between the first similarity feature.Wherein the first similarity feature can be similar value.
Step S520 removes the name entity in the input information to obtain remaining word, and calculates described surplus
The second similarity feature between remaining word and synonym and context vocabulary.
Specifically, the word in phrase is removed in input information user inputted, gets remaining some words
Or word, the similarity feature of the word in these words or word and adjacent phrase is calculated, these words or word and its synonym are also calculated
Similarity feature, two parts similarity feature combine to obtain the second similarity feature.
Step S530 generates high dimension vector according to the input information, wherein the high dimension vector is according to the input
It is generated in information with the presence or absence of default vocabulary.
Specifically, for the natural language question sentence of user's input, whether default vocabulary is appeared according to the word in the question sentence
In, to generate high dimension vector corresponding with the question sentence.Wherein each of high dimension vector position all represents a word, such as
There are the words in the fruit natural language question sentence, then value in the position is 1, it is otherwise 0.Such as user inputs " aspirin
It is which patient eats ", if only this four words of aspirin are present in default vocabulary, height corresponding to the question sentence
It is 1 that dimensional vector, which is exactly in the position that this four words of aspirin occur, other are 0, and the dimension of the high dimension vector can be according to reality
The setting of border situation.
Step S540 is generated according to the first similarity feature, the second similarity feature and the high dimension vector
Described eigenvector.
Specifically, the first similarity characteristic value, the second similarity feature and high dimension vector are spliced, to get
Final feature vector.
In the 6th embodiment, referring to FIG. 6, providing a kind of question and answer of Chinese knowledge mapping based on study sequence
Device 600.The question and answer system 600 includes:
First obtains module 610, for obtaining the input information of user.
Wherein, input information can be natural language retrieval sentence (such as question sentence), as user inputs question sentence on search website:
" what medicine cough needs to eat " the present embodiment inputs the mode of information without limitation to acquisition.
Identification and link module 620, the name entity in the input information for identification, and by the name chain of entities
It is connected to candidate entity corresponding with the name entity in the Chinese knowledge mapping, entity pair is formed, wherein the entity
To including the name entity and the candidate entity.
Specifically, by carrying out sequence labelling, then root to input information using mark set method, Recognition with Recurrent Neural Network model
The identification of name entity is completed according to the result of sequence labelling (specific steps will be discussed in detail in a second embodiment).Such as " cough
Cough and what medicine needed to eat ", first pass through BIO mark set method and it be labeled, the vector of the question sentence is obtained according to annotation results
Information, then using the vector information as the input of Recognition with Recurrent Neural Network model, thus this name entity that identifies " cough ".It connects
, which is corresponded to globally unique identifier (the Globally Unique in Chinese knowledge mapping
Identifier, GUID), thus by the name entity link into knowledge mapping it is corresponding candidate entity.In addition, knowledge mapping
Each of candidate entity uniquely correspond to a GUID, it is candidate to distinguish the difference in Chinese knowledge mapping by the GUID
Entity.
Wherein, Chinese knowledge mapping is a kind of new technique for storing labyrinth information.It is deposited in Chinese knowledge mapping
A host of facts type knowledge is stored up, relation information of the storage inside between entity and entity.Chinese knowledge mapping is mostly with RDF
The format memory data of (Resource Description Framework), a fact are represented as (S, P, an O) three
Tuple, shaped like (subject, predicate, object), wherein S and O is expressed as entity, and O is sometimes also indicated as attribute value, P
Indicate the relationship between S and O.Entity link be solve name entity ambiguity problem a kind of important method, this method pass through by
Have ambiguous entity denotion item and is linked to the elimination for realizing entity ambiguity in given knowledge mapping.
Matching module 630, for matching the candidate entity in the Chinese knowledge mapping by relational model
Candidate relationship.
Specifically, relationship templates are understood expressed by the input information (such as question sentence) of user by natural language understanding technology
Semanteme, and matched with the relationship P in the triple (S, P, O) in Chinese knowledge mapping, which is determined with this
Corresponding candidate relationship in the semantic and Chinese knowledge mapping of expression.Wherein, relationship templates are by Chinese knowledge mapping
Some triples are extracted, and extract relation information from these triples, to be obtained according to these relation informations by training
Relationship templates corresponding with these relation informations.
Form module 640, for according to the entity to the candidate relationship, form candidate triple;It is wherein described
Candidate triple includes the name entity, the candidate entity and the candidate relationship.
Specifically, the name entity identified using above-mentioned steps, the name entity are corresponding in Chinese knowledge mapping
Candidate entity and candidate relationship, to form each candidate triple.
Second acquisition module 650, for obtaining the corresponding sequence of each candidate's triple based on study order models
As a result.
Specifically, using each candidate triple as the input of study order models, by a system of the study order models
Column count, thus output ranking results corresponding with each candidate's triple.Wherein ranking results can it is more forward according to ranking more
Accurately, more more inaccurate rule is arranged ranking rearward, is also possible to other modes, the present embodiment does not limit.
Wherein, study order models are calculated using study sort algorithm.Learn sort algorithm (Learning
Torank, LTR) be a kind of supervised learning (SupervisedLearning, SL) sort method.LTR generally has three classes
Method: single document method (Pointwise), document is to method (Pairwise), lists of documents method (Listwise).This implementation
Study sort algorithm is using document to method (Pairwise) in example.
In one embodiment, study order models are to be formed by second by training first sample and each candidate triple
What sample obtained, wherein first sample is the triple being made of the model answer of inputted information.For example, according to a question sentence
Model answer triple (name entity, candidate entity, candidate relationship), take 10 times at random in Chinese knowledge mapping
Entity is selected, is acquiring candidate relationship according to this candidate entity, finally obtaining 50 triples, (name entity, is waited at candidate entity
Select relationship) composition negative sample (N).Wherein the triple (name entity, candidate entity, candidate relationship) of the model answer is positive
Sample (P).By a combination in the positive sample (P) and negative sample (N), generate two samples, be (P, N) sample and (N,
P) sample.The label of (P, N) sample is 1, and the label of (N, P) sample is 0.Learn order models can according to the sample training and
It arrives.
Third obtains module 660, described defeated to obtain for inquiring the Chinese knowledge mapping according to the ranking results
Enter the answer of information.
Specifically, according to the ranking results of each candidate triple, candidate triple of the ranking before preset value is chosen, then
These selected candidate triples are converted into the query language in Chinese knowledge mapping, to hold in Chinese knowledge mapping
The row query statement, returns to the corresponding answer of input information after inquiry.
In addition, referring to FIG. 7, the question and answer system 600 of the Chinese knowledge mapping based on study sequence further includes line lower module
700, which is used to prepare for the operation of above-mentioned question and answer system.
Line lower module 700 include entity referral rate unit 710, synonym collector unit 720, context excavate unit 730,
Question template unit 740 and study sequencing unit 750.
Entity referral rate unit 710 is given a mark for being mentioned number to the candidate entity in Chinese knowledge mapping.Tool
Body, referral rate marking is carried out to the candidate entity in Chinese knowledge mapping, wherein the referral rate indicates that candidate's entity is benefited from
The degree of concern at family.This part can by be ready for referral rate ranking (such as: the most concerned drug of patient is ranked
List), it can also be putd question to by crawling user on the network, the frequency that computational entity is referred to by user.
Synonym collector unit 720 is used to collect the relation name of each candidate relationship in Chinese knowledge mapping, wherein closing
It is the synonym that title includes title and title.
Specifically, each candidate relationship has a title, such as " xx drug treatment xx disease in Chinese knowledge mapping
This relationship of disease ", title is ... indication ..., but due to the diversity of Chinese natural language, user may
" what xx medicine cures mainly " said, " what xx medicine function is " etc..So needing to collect the same of relation name (or relationship predicate)
Adopted word.The synonym collector unit 612 is used to collect the relation name of each candidate relationship in Chinese knowledge mapping, this relationship
Title includes the synonym of title He the title, so that it is guaranteed that the accuracy of later period question and answer.
Context excavates unit 730 and is used to be based on text mining method, finds out two candidate entities in Chinese knowledge mapping
Between connection relationship.Specifically, context excavates unit based entirely on the text mining of remote supervisory.In two candidate entities
Between connection relationship (triple for considering that longest 2 is jumped is true) may have it is multiple.In the text set of professional domain, find this two
A word that a candidate's entity occurs simultaneously does the analysis of dependency grammar tree to the words, if the two entities are in dependency grammar
Minimum path length on tree is less than or equal to 4, then the word on this shortest path is just used as relationship between the two candidate entities (can
Can have multiple) cliction up and down (if this word is not the synonym of relationship).The general text information of professional domain is (as specially
Industry document etc.) data are abundant, but question and answer corpus (the question and answer corpus for being especially suitable for current knowledge mapping) may be relatively dilute
It lacks.By text mining, a large amount of language ambience informations can be provided for the question and answer system, to effectively utilize external resource.
Question template unit 740 is used to question sentence being divided into question sentence form predetermined.Specifically, by question sentence according to pre-
The question sentence form first defined is divided, and the search in Chinese knowledge mapping so is also more convenient, is more efficient.This step can be with
The relation space that regulation is compared is within jumping apart from main body entity double bounce or three.
Learn sequencing unit 750 to be used to obtain training data according to question sentence.Specifically, study sequencing unit is obtained according to question sentence
Training data is got, such a sort algorithm of pairwise learning to rank is based on.Although possible question and answer expect number
According to less, but training data can be expanded by way of generating negative sample, obtain the preferable Question-Answering Model of performance.
Wherein, referring to FIG. 8, synonym collector unit 720 include mark subelement 721, recording frequency subelement 722 and
Manual examination and verification subelement 723.
Mark subelement 721 is used to mark the relationship of candidate entity in the entity and knowledge mapping in question sentence.Recording frequency
Subelement 722 is used to remove entity name, stop words and the punctuation mark in question sentence, obtains remaining word, utilizes the inverse text of word frequency-
This frequency approach gives a mark to the remaining word, obtains the score value of remaining word, and record score value is more than the remaining word of preset value
Language.
Specifically, after recording frequency subelement 722 is by removing the entity name in question sentence, stop words and punctuation mark,
The remaining word in question sentence is got, using the inverse text frequency approach of word frequency-, gives a mark to the residue word, it is higher to collect score
Word, such as before score value ranking 15 word.
Wherein, TF-IDF is the abbreviation of Term Frequency-Inverse Document Frequency, i.e. " word frequency-
Inverse text frequency ".It consists of two parts, TF and IDF.What TF was indicated is word frequency, and text has namely been made in vectorization before
In each word frequency of occurrences statistics, and as text feature.IDF, i.e. " inverse text frequency ".The word frequency of some words is very high but again
The property wanted is but very low, and IDF is exactly to help us and react the importance of this word, and then to correct the word only indicated with word frequency special
Value indicative.
For summary, IDF reflects the frequency that a word occurs in all texts, if a word is in many texts
Middle appearance, then its IDF value should be low, such as " I " word.And in turn if a word occurs in fewer text,
So its IDF value should be high.Such as the noun such as " machine learning " of some professions.Such word IDF value should be high.One pole
The case where end, if a word all occurs in all texts, its IDF value should be 0.
Manual examination and verification subelement 823 falls synonym unreasonable under each relationship for manual filtration.Specifically, as wished
Synset is more accurate, can do certain manual examination and verification, i.e., under every class relationship, manual filtration falls unreasonable synonym.
The present invention also provides a kind of computer equipments, can such as execute smart phone, tablet computer, the notebook electricity of program
Brain, desktop computer, rack-mount server, blade server, tower server or Cabinet-type server (including independent clothes
Server cluster composed by business device or multiple servers) etc..The computer equipment of the present embodiment includes at least but unlimited
In: memory, the processor etc. of connection can be in communication with each other by device bus.
The present embodiment also provides a kind of computer readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory
(for example, SD or DX memory etc.), random access storage device (RAM), static random-access memory (SRAM), read-only memory
(ROM), electrically erasable programmable read-only memory (EEPROM), programmable read only memory (PROM), magnetic storage, magnetic
Disk, CD, server, App are stored thereon with computer program, phase are realized when program is executed by processor using store etc.
Answer function.The computer readable storage medium of the present embodiment is used for storage electronics 20, this hair is realized when being executed by processor
Bright answering method.
The serial number of the above embodiments of the invention is only for description, does not represent the advantages or disadvantages of the embodiments.
Through the above description of the embodiments, those skilled in the art can be understood that above-described embodiment side
Method can be realized by means of software and necessary general hardware platform, naturally it is also possible to by hardware, but in many cases
The former is more preferably embodiment.Based on this understanding, technical solution of the present invention substantially in other words does the prior art
The part contributed out can be embodied in the form of software products, which is stored in a storage medium
In (such as ROM/RAM, magnetic disk, CD), including some instructions are used so that a terminal device (can be mobile phone, computer, clothes
Business device, air conditioner or the network equipment etc.) execute method described in each embodiment of the present invention.
The above is only a preferred embodiment of the present invention, is not intended to limit the scope of the invention, all to utilize this hair
Equivalent structure or equivalent flow shift made by bright specification and accompanying drawing content is applied directly or indirectly in other relevant skills
Art field, is included within the scope of the present invention.
Claims (10)
1. a kind of answering method, which is characterized in that the answering method includes:
Obtain the input information of user;
Identify it is described input information in name entity, and by the name entity link to it is described Chinese knowledge mapping in institute
The corresponding candidate entity of name entity is stated, entity pair is formed, wherein the entity is to including the name entity and the time
Select entity;
The candidate relationship of the candidate entity is matched in the Chinese knowledge mapping by relationship templates;
According to the entity to the candidate relationship, candidate triple is formed, wherein candidate's triple includes the life
Name entity, the candidate entity and the candidate relationship;
The corresponding ranking results of each candidate's triple are obtained based on study order models;And
According to the ranking results inquiry Chinese knowledge mapping, to obtain the answer of the input information.
2. answering method as described in claim 1, which is characterized in that the name entity tool in the identification input information
Body includes:
The input information is labeled, annotation results are obtained;And according to the annotation results, pass through Recognition with Recurrent Neural Network mould
Type identifies the name entity in the input information.
3. answering method as described in claim 1, which is characterized in that the name entity in the identification input information,
And into the Chinese knowledge mapping, candidate entity corresponding with the name entity, formation are real by the name entity link
After the step of body pair, the answering method further include:
The similarity between name entity described in each entity pair and the candidate entity is calculated, wherein the similarity is
It is obtained according to chinese character similarity, pinyin character similarity, term vector similarity and the concerned degree of entity;
It is arranged the names with obtaining each entity to corresponding to each entity to sequence according to each similarity;And
The corresponding entity pair is chosen according to described arrange the names.
4. answering method as described in claim 1, which is characterized in that the relationship templates include first instance, second instance
And the relationship between first instance and second instance.
5. answering method as described in claim 1, which is characterized in that it is described based on study order models, obtain each time
The corresponding ranking results of triple are selected to specifically include:
Calculate the corresponding each feature vector of each triple;And
Each described eigenvector is input in the study order models to obtain the corresponding row of each candidate's triple
Sequence result.
6. answering method as claimed in claim 5, which is characterized in that each feature vector for calculating each triple
Step includes:
The first similarity feature between the name entity and the candidate entity is calculated according to the triple;
The name entity in the input information is removed to obtain remaining word, and calculates the remaining word and synonym
And the second similarity feature between context vocabulary;
High dimension vector is generated according to the input information, wherein the high dimension vector is whether there is according in the input information
What default vocabulary generated;And
According to the first similarity feature, the second similarity feature and the high dimension vector, described eigenvector is generated.
7. answering method as described in claim 1, which is characterized in that the study order models are by training first sample
It is formed by what the second sample obtained with each candidate triple, wherein first sample is answered by the standard of the input information
The triple that case is constituted.
8. a kind of question and answer system, which is characterized in that the question and answer system includes:
First obtains module, for obtaining the input information of user;
Identification and link module, the name entity in the input information for identification, and by the name entity link to institute
Candidate's entity corresponding with the name entity in Chinese knowledge mapping is stated, entity pair is formed, wherein the entity is to including
The name entity and the candidate entity;
Matching module is closed for matching the candidate of candidate entity in the Chinese knowledge mapping by relational model
System;
Form module, for according to the entity to the candidate relationship, form candidate triple;The wherein candidate ternary
Group includes the name entity, the candidate entity and the candidate relationship;
Second acquisition module, for obtaining the corresponding ranking results of each candidate's triple based on study order models;And
Third obtains module, for inquiring the Chinese knowledge mapping according to the ranking results, to obtain the input information
Answer.
9. a kind of computer equipment, can run on a memory and on a processor including memory, processor and storage
Computer program, which is characterized in that the processor realizes any one of claim 1 to 7 institute when executing the computer program
The step of stating answering method.
10. a kind of computer readable storage medium, is stored thereon with computer program, it is characterised in that: the computer program
The step of any one of claim 1 to 7 answering method is realized when being executed by processor.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910593110.6A CN110502621B (en) | 2019-07-03 | 2019-07-03 | Question answering method, question answering device, computer equipment and storage medium |
PCT/CN2020/093141 WO2021000676A1 (en) | 2019-07-03 | 2020-05-29 | Q&a method, q&a device, computer equipment and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910593110.6A CN110502621B (en) | 2019-07-03 | 2019-07-03 | Question answering method, question answering device, computer equipment and storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110502621A true CN110502621A (en) | 2019-11-26 |
CN110502621B CN110502621B (en) | 2023-06-13 |
Family
ID=68585335
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910593110.6A Active CN110502621B (en) | 2019-07-03 | 2019-07-03 | Question answering method, question answering device, computer equipment and storage medium |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN110502621B (en) |
WO (1) | WO2021000676A1 (en) |
Cited By (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111259653A (en) * | 2020-01-15 | 2020-06-09 | 重庆邮电大学 | Knowledge graph question-answering method, system and terminal based on entity relationship disambiguation |
CN111339269A (en) * | 2020-02-20 | 2020-06-26 | 来康科技有限责任公司 | Knowledge graph question-answer training and application service system with automatically generated template |
CN111353298A (en) * | 2020-02-17 | 2020-06-30 | 杭州网易再顾科技有限公司 | Character sequence generation method, device, equipment and computer readable storage medium |
CN111368048A (en) * | 2020-02-26 | 2020-07-03 | 京东方科技集团股份有限公司 | Information acquisition method and device, electronic equipment and computer readable storage medium |
CN111753055A (en) * | 2020-06-28 | 2020-10-09 | 中国银行股份有限公司 | Automatic client question and answer prompting method and device |
CN111883230A (en) * | 2019-12-18 | 2020-11-03 | 深圳数字生命研究院 | Method and device for generating diet data, storage medium and electronic device |
CN111950303A (en) * | 2020-10-19 | 2020-11-17 | 平安科技(深圳)有限公司 | Medical text translation method, device and storage medium |
CN112100356A (en) * | 2020-09-17 | 2020-12-18 | 武汉纺织大学 | Knowledge base question-answer entity linking method and system based on similarity |
CN112182178A (en) * | 2020-09-25 | 2021-01-05 | 北京字节跳动网络技术有限公司 | Intelligent question answering method, device, equipment and readable storage medium |
WO2021000676A1 (en) * | 2019-07-03 | 2021-01-07 | 平安科技(深圳)有限公司 | Q&a method, q&a device, computer equipment and storage medium |
CN112579752A (en) * | 2020-12-10 | 2021-03-30 | 上海明略人工智能(集团)有限公司 | Entity relationship extraction method and device, storage medium and electronic equipment |
CN112733508A (en) * | 2021-03-30 | 2021-04-30 | 中国电子技术标准化研究院 | Standard text labeling and standard map building method and device |
CN112925887A (en) * | 2019-12-05 | 2021-06-08 | 北京四维图新科技股份有限公司 | Interaction method and device, electronic equipment, storage medium and text recognition method |
CN112948569A (en) * | 2019-12-10 | 2021-06-11 | 中国石油天然气股份有限公司 | Method and device for pushing scientific workflow diagram version based on active knowledge graph |
WO2021159632A1 (en) * | 2020-02-13 | 2021-08-19 | 平安科技(深圳)有限公司 | Intelligent questioning and answering method and apparatus, computer device, and computer storage medium |
CN113312854A (en) * | 2021-07-19 | 2021-08-27 | 成都数之联科技有限公司 | Type selection recommendation method and device, electronic equipment and readable storage medium |
CN113361269A (en) * | 2021-06-11 | 2021-09-07 | 南京信息工程大学 | Method for text emotion classification |
CN113420160A (en) * | 2021-06-24 | 2021-09-21 | 竹间智能科技(上海)有限公司 | Data processing method and device |
CN113495964A (en) * | 2021-04-28 | 2021-10-12 | 中国科学技术大学 | Method, device and equipment for screening triples and readable storage medium |
WO2022088671A1 (en) * | 2020-10-29 | 2022-05-05 | 平安科技(深圳)有限公司 | Automated question answering method and apparatus, device, and storage medium |
CN114444505A (en) * | 2020-10-30 | 2022-05-06 | 北京金山数字娱乐科技有限公司 | Text processing method and device |
CN114510558A (en) * | 2022-01-26 | 2022-05-17 | 北京博瑞彤芸科技股份有限公司 | Question-answering method and system based on traditional Chinese medicine knowledge graph |
CN114781387A (en) * | 2022-06-20 | 2022-07-22 | 北京惠每云科技有限公司 | Medical named entity recognition method and device, electronic equipment and storage medium |
CN116127053A (en) * | 2023-02-14 | 2023-05-16 | 北京百度网讯科技有限公司 | Entity word disambiguation, knowledge graph generation and knowledge recommendation methods and devices |
Families Citing this family (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11526688B2 (en) * | 2020-04-16 | 2022-12-13 | International Business Machines Corporation | Discovering ranked domain relevant terms using knowledge |
CN112818031B (en) * | 2021-01-26 | 2023-10-27 | 国网江苏省电力有限公司营销服务中心 | Potential high-energy-consumption enterprise mining method, system and storage medium based on NLP Chinese word segmentation technology |
CN112749268A (en) * | 2021-01-30 | 2021-05-04 | 云知声智能科技股份有限公司 | FAQ system sequencing method, device and system based on hybrid strategy |
CN113157935B (en) * | 2021-03-16 | 2024-02-27 | 中国科学技术大学 | Entity alignment based on relation context and graph neural network system and method |
CN115129828A (en) * | 2021-03-25 | 2022-09-30 | 科沃斯商用机器人有限公司 | Human-computer interaction method and device, intelligent robot and storage medium |
CN113127626B (en) * | 2021-04-22 | 2024-04-30 | 广联达科技股份有限公司 | Recommendation method, device, equipment and readable storage medium based on knowledge graph |
CN113128231A (en) * | 2021-04-25 | 2021-07-16 | 深圳市慧择时代科技有限公司 | Data quality inspection method and device, storage medium and electronic equipment |
CN115376504A (en) * | 2021-05-19 | 2022-11-22 | 北京小米移动软件有限公司 | Voice interaction method and device for intelligent product and readable storage medium |
CN113505586A (en) * | 2021-06-07 | 2021-10-15 | 中电鸿信信息科技有限公司 | Seat-assisted question-answering method and system integrating semantic classification and knowledge graph |
CN113515630B (en) * | 2021-06-10 | 2024-04-09 | 深圳数联天下智能科技有限公司 | Triplet generation and verification method and device, electronic equipment and storage medium |
CN113377923B (en) * | 2021-06-25 | 2024-01-09 | 北京百度网讯科技有限公司 | Semantic retrieval method, apparatus, device, storage medium and computer program product |
CN113449119A (en) * | 2021-06-30 | 2021-09-28 | 珠海金山办公软件有限公司 | Method and device for constructing knowledge graph, electronic equipment and storage medium |
CN113590783B (en) * | 2021-07-28 | 2023-10-03 | 复旦大学 | NLP natural language processing-based traditional Chinese medicine health preserving intelligent question-answering system |
CN113704494B (en) * | 2021-08-27 | 2024-04-05 | 北京百度网讯科技有限公司 | Entity retrieval method, device, equipment and storage medium based on knowledge graph |
CN113761167B (en) * | 2021-09-09 | 2023-10-20 | 上海明略人工智能(集团)有限公司 | Session information extraction method, system, electronic equipment and storage medium |
CN113946651B (en) * | 2021-09-27 | 2024-05-10 | 盛景智能科技(嘉兴)有限公司 | Maintenance knowledge recommendation method and device, electronic equipment, medium and product |
CN116089587B (en) * | 2023-02-20 | 2024-03-01 | 星环信息科技(上海)股份有限公司 | Answer generation method, device, equipment and storage medium |
CN116955592B (en) * | 2023-07-21 | 2024-02-09 | 广州拓尔思大数据有限公司 | Data processing method and system based on visual reasoning result |
CN117194633A (en) * | 2023-09-12 | 2023-12-08 | 河海大学 | Dam emergency response knowledge question-answering system based on multi-level multipath and implementation method |
CN118132729A (en) * | 2024-04-28 | 2024-06-04 | 支付宝(杭州)信息技术有限公司 | Answer generation method and device based on medical knowledge graph |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107748757A (en) * | 2017-09-21 | 2018-03-02 | 北京航空航天大学 | A kind of answering method of knowledge based collection of illustrative plates |
CN107832400A (en) * | 2017-11-01 | 2018-03-23 | 山东大学 | A kind of method that location-based LSTM and CNN conjunctive models carry out relation classification |
US9985982B1 (en) * | 2015-12-21 | 2018-05-29 | Cisco Technology, Inc. | Method and apparatus for aggregating indicators of compromise for use in network security |
CN108345702A (en) * | 2018-04-10 | 2018-07-31 | 北京百度网讯科技有限公司 | Entity recommends method and apparatus |
US20180341866A1 (en) * | 2017-05-26 | 2018-11-29 | Baidu Online Network Technology (Beijing) Co., Ltd. | Method of building a sorting model, and application method and apparatus based on the model |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108427707B (en) * | 2018-01-23 | 2021-05-04 | 深圳市阿西莫夫科技有限公司 | Man-machine question and answer method, device, computer equipment and storage medium |
CN109241294A (en) * | 2018-08-29 | 2019-01-18 | 国信优易数据有限公司 | A kind of entity link method and device |
CN110502621B (en) * | 2019-07-03 | 2023-06-13 | 平安科技(深圳)有限公司 | Question answering method, question answering device, computer equipment and storage medium |
-
2019
- 2019-07-03 CN CN201910593110.6A patent/CN110502621B/en active Active
-
2020
- 2020-05-29 WO PCT/CN2020/093141 patent/WO2021000676A1/en active Application Filing
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9985982B1 (en) * | 2015-12-21 | 2018-05-29 | Cisco Technology, Inc. | Method and apparatus for aggregating indicators of compromise for use in network security |
US20180341866A1 (en) * | 2017-05-26 | 2018-11-29 | Baidu Online Network Technology (Beijing) Co., Ltd. | Method of building a sorting model, and application method and apparatus based on the model |
CN107748757A (en) * | 2017-09-21 | 2018-03-02 | 北京航空航天大学 | A kind of answering method of knowledge based collection of illustrative plates |
CN107832400A (en) * | 2017-11-01 | 2018-03-23 | 山东大学 | A kind of method that location-based LSTM and CNN conjunctive models carry out relation classification |
CN108345702A (en) * | 2018-04-10 | 2018-07-31 | 北京百度网讯科技有限公司 | Entity recommends method and apparatus |
Non-Patent Citations (1)
Title |
---|
官赛萍: "面向知识图谱的知识推理研究进展", 《软件学报》, vol. 29, no. 10, pages 2966 - 2994 * |
Cited By (35)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2021000676A1 (en) * | 2019-07-03 | 2021-01-07 | 平安科技(深圳)有限公司 | Q&a method, q&a device, computer equipment and storage medium |
CN112925887A (en) * | 2019-12-05 | 2021-06-08 | 北京四维图新科技股份有限公司 | Interaction method and device, electronic equipment, storage medium and text recognition method |
CN112948569A (en) * | 2019-12-10 | 2021-06-11 | 中国石油天然气股份有限公司 | Method and device for pushing scientific workflow diagram version based on active knowledge graph |
CN111883230A (en) * | 2019-12-18 | 2020-11-03 | 深圳数字生命研究院 | Method and device for generating diet data, storage medium and electronic device |
CN111883230B (en) * | 2019-12-18 | 2024-05-07 | 深圳数字生命研究院 | Diet data generation method and device, storage medium and electronic device |
CN111259653A (en) * | 2020-01-15 | 2020-06-09 | 重庆邮电大学 | Knowledge graph question-answering method, system and terminal based on entity relationship disambiguation |
CN111259653B (en) * | 2020-01-15 | 2022-06-24 | 重庆邮电大学 | Knowledge graph question-answering method, system and terminal based on entity relationship disambiguation |
WO2021159632A1 (en) * | 2020-02-13 | 2021-08-19 | 平安科技(深圳)有限公司 | Intelligent questioning and answering method and apparatus, computer device, and computer storage medium |
CN111353298A (en) * | 2020-02-17 | 2020-06-30 | 杭州网易再顾科技有限公司 | Character sequence generation method, device, equipment and computer readable storage medium |
CN111339269A (en) * | 2020-02-20 | 2020-06-26 | 来康科技有限责任公司 | Knowledge graph question-answer training and application service system with automatically generated template |
CN111339269B (en) * | 2020-02-20 | 2023-09-26 | 来康科技有限责任公司 | Knowledge graph question-answering training and application service system capable of automatically generating templates |
CN111368048B (en) * | 2020-02-26 | 2024-06-25 | 京东方科技集团股份有限公司 | Information acquisition method, information acquisition device, electronic equipment and computer readable storage medium |
CN111368048A (en) * | 2020-02-26 | 2020-07-03 | 京东方科技集团股份有限公司 | Information acquisition method and device, electronic equipment and computer readable storage medium |
WO2021169718A1 (en) * | 2020-02-26 | 2021-09-02 | 京东方科技集团股份有限公司 | Information acquisition method and apparatus, electronic device, and computer-readable storage medium |
CN111753055A (en) * | 2020-06-28 | 2020-10-09 | 中国银行股份有限公司 | Automatic client question and answer prompting method and device |
CN111753055B (en) * | 2020-06-28 | 2024-01-26 | 中国银行股份有限公司 | Automatic prompt method and device for customer questions and answers |
CN112100356A (en) * | 2020-09-17 | 2020-12-18 | 武汉纺织大学 | Knowledge base question-answer entity linking method and system based on similarity |
CN112182178A (en) * | 2020-09-25 | 2021-01-05 | 北京字节跳动网络技术有限公司 | Intelligent question answering method, device, equipment and readable storage medium |
WO2021179693A1 (en) * | 2020-10-19 | 2021-09-16 | 平安科技(深圳)有限公司 | Medical text translation method and device, and storage medium |
CN111950303A (en) * | 2020-10-19 | 2020-11-17 | 平安科技(深圳)有限公司 | Medical text translation method, device and storage medium |
WO2022088671A1 (en) * | 2020-10-29 | 2022-05-05 | 平安科技(深圳)有限公司 | Automated question answering method and apparatus, device, and storage medium |
CN114444505A (en) * | 2020-10-30 | 2022-05-06 | 北京金山数字娱乐科技有限公司 | Text processing method and device |
CN112579752A (en) * | 2020-12-10 | 2021-03-30 | 上海明略人工智能(集团)有限公司 | Entity relationship extraction method and device, storage medium and electronic equipment |
CN112733508A (en) * | 2021-03-30 | 2021-04-30 | 中国电子技术标准化研究院 | Standard text labeling and standard map building method and device |
CN113495964A (en) * | 2021-04-28 | 2021-10-12 | 中国科学技术大学 | Method, device and equipment for screening triples and readable storage medium |
CN113495964B (en) * | 2021-04-28 | 2024-02-23 | 中国科学技术大学 | Triad screening method, device, equipment and readable storage medium |
CN113361269A (en) * | 2021-06-11 | 2021-09-07 | 南京信息工程大学 | Method for text emotion classification |
CN113361269B (en) * | 2021-06-11 | 2023-07-18 | 南京信息工程大学 | Text emotion classification method |
CN113420160A (en) * | 2021-06-24 | 2021-09-21 | 竹间智能科技(上海)有限公司 | Data processing method and device |
CN113312854B (en) * | 2021-07-19 | 2021-11-02 | 成都数之联科技有限公司 | Type selection recommendation method and device, electronic equipment and readable storage medium |
CN113312854A (en) * | 2021-07-19 | 2021-08-27 | 成都数之联科技有限公司 | Type selection recommendation method and device, electronic equipment and readable storage medium |
CN114510558A (en) * | 2022-01-26 | 2022-05-17 | 北京博瑞彤芸科技股份有限公司 | Question-answering method and system based on traditional Chinese medicine knowledge graph |
CN114781387A (en) * | 2022-06-20 | 2022-07-22 | 北京惠每云科技有限公司 | Medical named entity recognition method and device, electronic equipment and storage medium |
CN116127053B (en) * | 2023-02-14 | 2024-01-02 | 北京百度网讯科技有限公司 | Entity word disambiguation, knowledge graph generation and knowledge recommendation methods and devices |
CN116127053A (en) * | 2023-02-14 | 2023-05-16 | 北京百度网讯科技有限公司 | Entity word disambiguation, knowledge graph generation and knowledge recommendation methods and devices |
Also Published As
Publication number | Publication date |
---|---|
CN110502621B (en) | 2023-06-13 |
WO2021000676A1 (en) | 2021-01-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110502621A (en) | Answering method, question and answer system, computer equipment and storage medium | |
CN110399457B (en) | Intelligent question answering method and system | |
CN110633409B (en) | Automobile news event extraction method integrating rules and deep learning | |
Lerman et al. | Using the structure of web sites for automatic segmentation of tables | |
CN110990590A (en) | Dynamic financial knowledge map construction method based on reinforcement learning and transfer learning | |
CN107180045B (en) | Method for extracting geographic entity relation contained in internet text | |
CN110263180B (en) | Intention knowledge graph generation method, intention identification method and device | |
Kowalski | Information retrieval architecture and algorithms | |
CN110298033A (en) | Keyword corpus labeling trains extracting tool | |
CN112819023B (en) | Sample set acquisition method, device, computer equipment and storage medium | |
CN105205699A (en) | User label and hotel label matching method and device based on hotel comments | |
CN110188197B (en) | Active learning method and device for labeling platform | |
CN109471949B (en) | Semi-automatic construction method of pet knowledge graph | |
CN104102721A (en) | Method and device for recommending information | |
CN111858940B (en) | Multi-head attention-based legal case similarity calculation method and system | |
CN105393265A (en) | Active featuring in computer-human interactive learning | |
CN106708929B (en) | Video program searching method and device | |
CN102663129A (en) | Medical field deep question and answer method and medical retrieval system | |
CN113535917A (en) | Intelligent question-answering method and system based on travel knowledge map | |
CN103886020B (en) | A kind of real estate information method for fast searching | |
CN102955848A (en) | Semantic-based three-dimensional model retrieval system and method | |
CN112559684A (en) | Keyword extraction and information retrieval method | |
CN112328800A (en) | System and method for automatically generating programming specification question answers | |
Noel et al. | Applicability of Latent Dirichlet Allocation to multi-disk search | |
CN116881436A (en) | Knowledge graph-based document retrieval method, system, terminal 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 |