CN109388793A - Entity mask method, intension recognizing method and corresponding intrument, computer storage medium - Google Patents
Entity mask method, intension recognizing method and corresponding intrument, computer storage medium Download PDFInfo
- Publication number
- CN109388793A CN109388793A CN201710655187.2A CN201710655187A CN109388793A CN 109388793 A CN109388793 A CN 109388793A CN 201710655187 A CN201710655187 A CN 201710655187A CN 109388793 A CN109388793 A CN 109388793A
- Authority
- CN
- China
- Prior art keywords
- sentence
- vector
- word
- entity
- partly
- 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
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/237—Lexical tools
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
- G06F40/295—Named entity recognition
Abstract
The present invention provides a kind of entity mask method, intension recognizing method and corresponding intruments, computer storage medium.Wherein entity mask method includes: the first expression vector for obtaining at least partly word at least partly the attribute tags of word carry out Chinese word coding in sentence using knowledge mapping；Based on sentence structure at least partly word carries out Chinese word coding in the sentence, the second expression vector of at least partly word is obtained；First expression vector sum second is expressed vector to merge, obtains the entity annotation results to the sentence.Intension recognizing method includes: to be combined coding using attribute tags of the knowledge mapping at least partly word in sentence, obtains first vector of the sentence；The sentence is encoded based on sentence structure, obtains second vector of the sentence；First vector sum, second vector of the sentence is merged, the intention assessment result to the sentence is obtained.Mode provided by the invention can be improved the accuracy of entity mark and intention assessment.
Description
[technical field]
The present invention relates to computer application technology, in particular to a kind of entity mask method, intension recognizing method and
Corresponding intrument, computer storage medium.
[background technique]
Natural language processing is important or even core a part of artificial intelligence, and the purpose is to understand in short to want table
It mainly include two main tasks up to what: entity mark and intention assessment.Wherein entity mark is entity word in mark a word
Attribute tags, it is intended that identification is identification in short wants that intention or purpose realized.For example, if there is so in short
" which film Zhou Jielun drilled ", the task of entity mark are that entity word " Zhou Jielun " is labeled as Movie_actor label,
Movie_actor refers to video display performer；And intention assessment is to identify that word is to obtain which film a performer drilled.
Current entity mark and intension recognizing method be all based only on sentence structure, this sentence structure that is based purely on
Mode often will cause the problems such as intention assessment and low entity mark accuracy rate.
[summary of the invention]
In view of this, the present invention provides a kind of entity mask method, intension recognizing method and corresponding intrument, computers to deposit
Storage media, in order to improve the accuracy rate of entity mark and intention assessment.
Specific technical solution is as follows:
The present invention provides a kind of entity mask methods, this method comprises:
Using knowledge mapping at least partly the attribute tags of word carry out Chinese word coding in sentence, at least partly word is obtained
First expression vector；
Based on sentence structure at least partly word carries out Chinese word coding in the sentence, the second of at least partly word is obtained
Express vector；
First expression vector sum second is expressed vector to merge, obtains the entity annotation results to the sentence.
The present invention also provides a kind of intension recognizing methods, this method comprises:
It is combined coding using attribute tags of the knowledge mapping at least partly word in sentence, obtains the sentence
First vector；
The sentence is encoded based on sentence structure, obtains second vector of the sentence；
First vector sum, second vector of the sentence is merged, the intention assessment knot to the sentence is obtained
Fruit.
The present invention provides a kind of entity annotation equipments, which is characterized in that the device includes:
First Chinese word coding unit, for using knowledge mapping in sentence at least partly word attribute tags carry out word volume
Code obtains the first expression vector of at least partly word；
Second Chinese word coding unit, for, at least partly word carries out Chinese word coding in the sentence, being obtained based on sentence structure
Second at least partly word expresses vector；
Vector Fusion unit is merged for the first expression vector sum second to be expressed vector, is obtained to the sentence
Entity annotation results.
The present invention also provides a kind of intention assessment device, which includes:
First coding unit, for being combined using attribute tags of the knowledge mapping at least partly word in sentence
Coding, obtains first vector of the sentence；
Second coding unit obtains the second of the sentence for encoding based on sentence structure to the sentence
Sentence vector；
Vector Fusion unit is obtained for merging first vector sum, second vector of the sentence to institute
State the intention assessment result of sentence.
The present invention provides a kind of equipment, including
Memory, including one or more program；
One or more processor is coupled to the memory, executes one or more of programs, on realizing
State the operation executed in method.
The present invention also provides a kind of computer storage medium, the computer storage medium is encoded with computer journey
Sequence, described program by one or more computers when being executed, so that one or more of computers execute in the above method
The operation of execution.
As can be seen from the above technical solutions, knowledge mapping is introduced entity mark and intention assessment by the present invention, that is, is passed through
Entity attributes information in knowledge mapping is merged with the mode based on sentence structure, Lai Jinhang entity mark and intention are known
Not, the mode that the prior art that compares is based purely on sentence structure improves accuracy.
[Detailed description of the invention]
Fig. 1 is the method flow diagram of entity provided in an embodiment of the present invention mark；
Fig. 2 is the schematic diagram provided in an embodiment of the present invention that Chinese word coding is carried out using knowledge mapping；
Fig. 3 is the schematic diagram provided in an embodiment of the present invention that Chinese word coding is carried out based on sentence structure；
Fig. 4 is the signal that fusion knowledge mapping provided in an embodiment of the present invention and sentence structure mode carry out entity mark
Figure；
Fig. 5 is the method flow diagram of intention assessment provided in an embodiment of the present invention；
Fig. 6 is the schematic diagram provided in an embodiment of the present invention that sentence coding is carried out using knowledge mapping；
Fig. 7 is the signal that fusion knowledge mapping provided in an embodiment of the present invention and sentence structure mode carry out intention assessment
Figure；
Fig. 8 is the structure chart of entity annotation equipment provided in an embodiment of the present invention；
Fig. 9 is intention assessment structure drawing of device provided in an embodiment of the present invention；
Figure 10 is the structure chart of example apparatus provided in an embodiment of the present invention.
[specific embodiment]
To make the objectives, technical solutions, and advantages of the present invention clearer, right in the following with reference to the drawings and specific embodiments
The present invention is described in detail.
Core of the invention thought is, knowledge mapping is introduced entity mark and intention assessment, i.e., will be in knowledge mapping
Entity attributes information is merged with the mode based on sentence structure, Lai Jinhang entity mark and intention assessment, to improve
Accuracy.Method and apparatus provided by the invention are described in detail respectively below with reference to embodiment.
Fig. 1 is the method flow diagram of entity provided in an embodiment of the present invention mark, and as shown in fig. 1, this method can wrap
Include following steps:
In 101, knowledge mapping is pre-processed.
In knowledge mapping, the relationship that is stored between each entity and the corresponding attribute information of each entity, each entity.But know
Knowing map is usually with field/category division, for example, entity " Zhou Jielun " is corresponding with attribute mark in music field/classification
Label: " singer ", " composer " and " word author ", while in video display field/classification, there is also entity " Zhou Jielun ", correspond to
There is attribute tags " performer ".Utilization to knowledge mapping for convenience in the embodiment of the present invention, can first to knowledge mapping into
Row pretreatment.Specifically, it may comprise steps of:
S11, attribute tags of the entity each in knowledge mapping in each field are integrated first, it is corresponding obtains each entity
Attribute tags.
Still by taking above-mentioned entity " Zhou Jielun " as an example, by it after the attribute tags in each field are integrated respectively, obtain
The corresponding all properties label of entity " Zhou Jielun " are as follows: " singer ", " composer ", " word author ", " performer ".
S12, the corresponding attribute tags of each entity are stored in key assignments storage engines.
After obtaining the corresponding attribute tags of each entity, respectively using entity as key (key), by the corresponding attribute mark of entity
Label are as value (value), then by each key assignments (key-value) to being stored in key assignments storage engines.
It should be noted that the purpose of above-mentioned pretreatment to knowledge mapping is to entity for subsequent convenience in knowledge mapping
In attribute tags quickly searched, but be not the present invention it is had to carry out the step of.It is of course also possible to use other are right
Knowledge mapping carries out pretreated mode.
In 102, Chinese word coding is carried out to the attribute tags of word each in sentence using knowledge mapping, obtains the of each word
One expression vector.
In this step, the first expression vector of each word obtained using knowledge mapping, its purpose is to allow each word
The first expression vector in include entity attributes information in knowledge mapping.Specifically, it can be realized by following steps:
S21, the entity and the corresponding attribute tags of the entity in knowledge mapping identification sentence are utilized.
Sentence can be matched using longest match principle in knowledge mapping, be identified in sentence in this step
Entity.Specifically, each n-gram (n-gram) of sentence is obtained, in embodiments of the present invention, n-gram refers to continuous n
The collocation that word is constituted, each value that wherein n is 1 or more.Each n-gram is matched with knowledge mapping respectively, which n- seen
Gram is matched to entity in knowledge mapping, when there is the multiple n-gram to overlap to be all matched to entity, takes wherein length
Longest n-gram is as the entity identified.
For example, respectively obtaining each n-gram includes: in " which film Zhou Jielun drilled " the words
1-gram: " week ", " outstanding person ", " human relations " ..., " shadow "；
2-gram: " Zhou Jie ", " Jie Lun ", " human relations are drilled " ..., " film "；
3-gram: " Zhou Jielun ", " Jie Lun is drilled ", " human relations were drilled " ..., " a little films "；
……
Wherein " Zhou Jie " can be matched to entity in knowledge mapping, and " Zhou Jielun " can be also matched in knowledge mapping
Entity, and there is overlapping in the two, then taking length longest " Zhou Jielun " as the entity identified.
When determining the corresponding attribute tags of each entity, key assignments storage engines can be inquired, are searched using entity as key pairs
The value answered.
S22, sentence is segmented using recognition result, attribute tags is marked to obtained each word.
When segmenting to sentence, the entity that will identify that is as independent word, then to the other content of sentence
It is segmented.Which still by taking " which film Zhou Jielun drilled " as an example, after being segmented, obtain: " Zhou Jielun ", " drilling ", " mistake ", "
A bit ", " film ".
After marking attribute tags to each word, " Zhou Jielun " mark " singer ", " composer ", " word author ", " performer ", by
The entity of knowledge mapping is not it in " drilling ", " mistake ", " which ", " film ", therefore it is corresponding to mark " O " instruction
Attribute tags.
It should be noted that being described by taking " each word " in sentence as an example in embodiments of the present invention, but also not
It excludes to be handled using " at least partly word " in sentence.For example, in this step, after being segmented to sentence, to
At least partly word mark attribute tags arrived, such as only mark attribute tags to the entity in instruction map.
S23, Chinese word coding is carried out to the attribute tags of each word, and coding result is carried out to the conversion of full articulamentum, obtained
First expression vector of each word.
In this step, Chinese word coding is carried out to the attribute tags of each word, it is therefore an objective to turn the attribute tags set of each word
It is changed to the coding that a string of computers can identify.The coding mode used in the present embodiment can include but is not limited to one-hot
(solely heat) coding.
As shown in Fig. 2, respectively obtaining one after the corresponding attribute tags set of each word carries out one-hot coding respectively
A coding result.The length of the coding result can be the total quantity of attribute tags, such as there are M attribute marks in knowledge mapping
Label, then coding result is exactly M, each corresponds to an attribute tags.Everybody value is to be shown to be in coding result
It is no that there are attribute tags corresponding to this.Such as " Zhou Jielun " is carried out in the result of Chinese word coding, having 4 is 1, shows " Zhou Jie
There are this corresponding attribute tags in 4 positions for human relations ".
For the coding result of one-hot, the conversion of full articulamentum is carried out, it is therefore an objective to by the volume of the attribute tags of each word
Code result maps on entity tag, which is exactly the label that entity mark is carried out to word in sentence.By connecting entirely
After connecing layer conversion, the first expression vector of each word is obtained.
In embodiments of the present invention, above-mentioned full articulamentum can be trained in advance.Training process may include: preparatory
Using the sentence for having been marked with entity tag as training sample, above-mentioned reality is carried out to the sentence in training sample using knowledge mapping
After body identification, participle, the mark of attribute tags, one-hot coding, as the input of the full articulamentum, each word in the sentence
The first expression vector that corresponding entity tag is constituted is that the target of the full articulamentum exports, and is trained to full articulamentum.Instruction
The full articulamentum got is actually to carry out after one-hot coding coding result to the mapping of entity tag.
Continue as shown in Fig. 2, the corresponding one-hot coding result of each word respectively by full articulamentum conversion after, obtain
First expression vector of each word, respectively indicates are as follows: T-dict1, T-dict2, T-dict3, T-dict4 and T-dict5.
In 103, based on sentence structure to word each in sentence carry out Chinese word coding, obtain the second of each word express to
Amount.
In this step, can specifically following steps be executed:
S31, the term vector for determining each word in sentence.
In determining sentence when the term vector of each word, existing term vector Core Generator can be used, such as
Word2vec etc. then can be for each word point using the word2vec based on word2vec is trained in advance based on semantic
Not Sheng Cheng term vector, the corresponding term vector length of each word is identical.The mode of this determining term vector be it is semantic-based, can
So that the distance between term vector embodies the correlation degree between phrase semantic, the higher word of correlation degree, right between semanteme
The distance between term vector answered is smaller.Current existing technology can be used in view of semantic-based term vector method of determination,
It is not described here in detail.
S32, the neural network for training term vector input in advance respectively obtain the second expression vector of each word.
By each term vector input neural network trained in advance, it is therefore an objective to be encoded to sentence according to word granularity.On
Such as two-way RNN (Recognition with Recurrent Neural Network), unidirectional RNN, CNN (convolutional neural networks) etc. can be used by stating neural network.Its
In preferably bidirectional RNN sentence is encoded because two-way RNN can be recycled.The basic thought of two-way RNN is to propose each
Training sequence is forwardly and rearwardly two RNN respectively, and the two RNN are connected to an output layer.This structure can incite somebody to action
The front and back contextual information that each in list entries is put is supplied to output layer.Specific to the present invention, when one participle of input
Good sentence, it is assumed that include n word in the sentence, then after two-way RNN, it will have n output vector, each word is corresponding
One vector.Due to the Memorability of RNN, i-th of vector contains the information of all words in front, therefore the output of the last one word
Vector is also referred to as " sentence vector ", because theoretically it contains the information of all words in front.
Still by taking " which film Zhou Jielun drilled " as an example, as shown in figure 3, each word " Zhou Jielun " obtained after participle,
After " drilling ", " mistake ", " which ", " film " determine corresponding term vector respectively, term vector is inputted into two-way RNN, so respectively
Obtain the lyrics and second expression vector, be denoted as respectively: output1, output2, output3, output4, output5, often
Second expression vector of a word all contains contextual information, i.e., divides the influence for considering sentence structure again, contain sentence
Structural information.Wherein output5 contains the information of entire sentence, can be referred to as sentence vector.
It should be noted that knowledge based map and the processing carried out based on sentence structure can in above-mentioned steps 102 and 103
Successively to execute in any order, also may be performed simultaneously.It is sequentially only one of side of execution shown in the present embodiment
Formula.
In 104, the first expression vector sum second is expressed vector to merge, obtains the entity mark knot to sentence
Fruit.
Expressing the fusion of vector to the first expression vector sum second in this step is actually to obtain to knowledge based map
To entity mark and the entity mark that is obtained based on sentence structure merged.Specifically, following steps can specifically be executed:
S41, the first of each word the expression vector sum second is expressed into vector respectively splice, obtain the third of each word
Express vector.
In this step, two vectors can be spliced according to preset sequence, so that a longer vector is obtained,
The vector is that third expresses vector.
It should be noted that may be used also other than the first expression vector sum second is expressed the mode that vector splices
Other amalgamation modes such as to be overlapped using the first expression vector sum second is expressed vector, but since the mode of splicing can
The influence of knowledge based map and the influence based on sentence structure are separately considered, thus in the conversion process of subsequent full articulamentum
The mode that different parameters is respectively adopted, therefore preferably splices.
S42, the result vector that the third expression vector of each word is converted to each word by full articulamentum.
The full articulamentum of the third expression vector input training in advance of each word is converted, so that each third be expressed
DUAL PROBLEMS OF VECTOR MAPPING obtains result vector after conversion to entity tag.Wherein the length of result vector is correspondent entity label
Total quantity, every corresponding each entity tag of result vector, everybody value correspond to the score of each entity tag.
In embodiments of the present invention, above-mentioned full articulamentum can be trained in advance.Training process may include: preparatory
Using the sentence for having been marked with entity tag as training sample, the step in above-mentioned steps 102 and 103 is executed respectively, that is, is directed to
The first expression vector sum second that sentence in training sample respectively obtains each word expresses vector, then expresses vector for first
Input with the second expression spliced result of vector (i.e. third expression vector) as the full articulamentum, the entity mark of the sentence
It signs and is trained as the output of full articulamentum.The full articulamentum that training obtains is expressed to carry out the third of each word in sentence
Vector to entity tag mapping.
S43, according to each word result vector to sentence carry out entity mark.
Each word is corresponding with a result vector, can select according to the score of entity tag each in result vector
The entity tag of highest scoring carries out entity mark to each word in sentence.
Still by taking " which film Zhou Jielun drilled " as an example, as shown in figure 4, expressing the second table of vector sum for the first of each word
Spliced respectively up to vector, obtains third expression vector.In Fig. 4, the first expression vector T-dict1 and second of " Zhou Jielun "
After expression vector Output1 is spliced, third expression vector K1 is obtained, other words are similar.Then by the third of each word
Expression vector K1, K2 ... K5 inputs full articulamentum respectively, respectively obtains the result vector of each word.Word " Zhou Jielun " is corresponding
Result vector in entity tag " Actor_name (actor names) " highest scoring, can using " Actor_name " to word
Language " Zhou Jielun " is labeled, and in the corresponding result vector of other words, the entity tag of highest scoring is " O ", and instruction is not
Entity, therefore other words are labeled using entity tag " O ".
Fig. 5 is the method flow diagram of intention assessment provided in an embodiment of the present invention, as shown in figure 5, this method may include
Following steps:
In 501, it is combined coding using attribute tags of the knowledge mapping to word each in sentence, obtains the of sentence
One vector.
Similarly with entity mark, knowledge mapping can be pre-processed first before this step, pretreated mistake
Cheng Buzai is described in detail, and may refer in Fig. 1 101 associated description.
In this step, first vector of sentence is obtained using knowledge mapping, in order to allow in first vector and wrap
Containing entity attributes information in knowledge mapping.Specifically, it can be realized by following steps:
S51, the entity and the corresponding attribute tags of the entity in knowledge mapping identification sentence are utilized.
In the embodiment shown in Figure 1 of realization in detail of this step, the step S21 in 102, details are not described herein.
S52, the attribute tags of each word are combined with coding, and coding result is carried out to the conversion of full articulamentum, obtained
To first vector of sentence.
It is unified that coding is combined to the attribute tags of each word in obtaining sentence after the attribute tags of each word, it obtains
To a coding result.The coding result is a vector, and the length of the vector corresponds to the total quantity of attribute tags, each is right
An attribute tags are answered, everybody value is weight of the attribute tags in sentence.
It wherein, can be according to time that attribute tags in the sentence occur when determining weight of the attribute tags in sentence
It counts and is determined with the corresponding identical entity attributes number of labels of the attribute tags.Specifically, attribute tags labeliWeightIt can be determined using following formula:
Wherein, m indicates that m-th of word in sentence, M indicate the number of word in sentence.aimIndicate label labeliIt is right
In the value of m-th of word, if labeliIt is not the attribute tags of m word, then aimValue take 0, if labeliIt is m-th of word
The attribute tags of language, then aimValue beWherein count (labelm) be m-th of word attribute tags
Number.
Still by taking sentence " which film Zhou Jielun drilled " as an example, " Zhou Jielun " corresponding all properties label are as follows: " singer ",
Corresponding attribute tags are not present in " composer ", " word author ", " performer ", other words in knowledge mapping.So for
For " singer " this attribute tags, the weight in sentence are as follows:So in coding result
In, " singer " corresponding position value is 0.25.Similarly, " composer ", " word author ", " performer " respectively correspond in coding result
Position value be 0.25, and in coding result the corresponding position value of other attribute tags be 0.
After obtaining above-mentioned coding result, coding result is converted by full articulamentum, it is therefore an objective to by sentence based on category
The coding result of property label maps on entity tag.The entity tag is exactly the mark that entity mark is carried out to word in sentence
Label.After the conversion of full articulamentum, first vector of sentence is obtained.The length correspondent entity label of first vector it is total
Quantity, every value of first vector are weight of the corresponding entity tag of this in sentence.
In embodiments of the present invention, above-mentioned full articulamentum can be trained in advance.Training process may include: preparatory
Using the sentence for having been marked with entity tag as training sample, above-mentioned reality is carried out to the sentence in training sample using knowledge mapping
After body identification, participle, the mark of attribute tags, assembly coding, input of the obtained coding result as the full articulamentum, the sentence
First vector that the corresponding entity tag of each word is constituted in son is exported as the target of the full articulamentum, to full articulamentum into
Row training.The obtained full articulamentum of training is actually to be combined after coding coding result to the mapping of entity tag.
Process in this step can be as shown in fig. 6, the attribute tags of each word pass through in " which film Zhou Jielun drills "
After crossing assembly coding, obtained coding result finally obtains first vector, is expressed as S-dict by full articulamentum.
In 502, sentence is encoded based on sentence structure, obtains second vector of sentence.
In this step, can specifically following steps be executed:
S61, the term vector for determining each word in sentence.
S62, the neural network for training the term vector input of each word in advance, obtain second vector of sentence.
Specifically, by after the term vector input of each word neural network trained in advance, the second of each word is respectively obtained
Vector is expressed, using the second expression vector of the last one word as second vector of sentence.
The term vector of each word in above-mentioned determining sentence, and the nerve net that the term vector input of each word is trained in advance
The treatment process of network is corresponding in step 103 in embodiment illustrated in fig. 1 to realize that unanimously details are not described herein.Only obtaining
After second expression vector of each word, using the second expression vector of the last one word as second vector of sentence, and its
He in sentence intention assessment and is not used the second expression vector of word.That is, the output5 used in Fig. 3 makees
For second vector of the sentence.
In 503, first vector sum, second vector of sentence is merged, the intention assessment knot to sentence is obtained
Fruit.
It is actually to be obtained to knowledge based map to the fusion of first vector sum, second vector in this step
Intent information and the intent information obtained based on sentence structure are merged.The wherein entity annotation results pair of knowledge based map
There is intention assessment very big influence to be correctly labeled as " Zhou Jielun " still by taking " which film Zhou Jielun drilled " as an example
" performer " is very big on correctly intention assessment result " which film a performer drilled " influence, if by entity " Zhou Jielun " mistake
Ground is labeled as " singer ", then is just likely to be unable to get above-mentioned intention assessment result.
Specifically, this step may comprise steps of:
S71, first vector sum, second vector is spliced, obtains third sentence vector.
In this step, two vectors can be spliced according to preset sequence, so that a longer vector is obtained,
The vector is third sentence vector.
It should be noted that can also be adopted other than the mode for splicing first vector sum, second vector
With first vector sum, second vector other amalgamation modes such as to be overlapped to.But consider since the mode of splicing can separate
The influence of knowledge based map and influence based on sentence structure, to be respectively adopted in the conversion process of subsequent full articulamentum
Different parameters, therefore the mode preferably spliced.
S72, third sentence vector is converted into result vector by full articulamentum.
The full articulamentum of third sentence vector input training in advance is converted, thus by third sentence DUAL PROBLEMS OF VECTOR MAPPING to sentence
It is intended to, obtains result vector after conversion.The length of the result vector corresponds to the categorical measure of sentence intention, result vector
The score that everybody corresponding all kinds of sentences are intended to.
In embodiments of the present invention, above-mentioned full articulamentum can be trained in advance.Training process may include: preparatory
It will have determined that the sentence that sentence is intended to executes the step in above-mentioned steps 501 and 502 as training sample respectively, and for instruction
The sentence practiced in sample respectively obtains first vector sum, second vector, then by first vector sum, second vector splicing
Input of the result (i.e. third sentence vector) as full articulamentum afterwards, the sentence of the sentence are intended to the output as full articulamentum,
It is trained.The mapping that the full articulamentum that training obtains is intended to carry out the third sentence vector of sentence to sentence.
S73, determine that sentence is intended to according to result vector.
In this step, the score value of classification can be intended to according to sentence each in result vector to determine that sentence is intended to, such as
The highest sentence of score value is intended to the intention as the sentence identified.
Still by taking " which film Zhou Jielun drilled " as an example, as shown in fig. 7, by first vector S-dict of sentence and second
After sentence vector Output5 is spliced, third sentence vector K is obtained.Then third sentence vector K is inputted into full articulamentum, finally
To a result vector.The sentence of highest score is intended in the result vector are as follows: " which film a performer drilled ".
It is the detailed description carried out to method provided by the present invention above, below with reference to embodiment to dress provided by the invention
It sets and is described in detail.
Fig. 8 is the structure chart of entity annotation equipment provided in an embodiment of the present invention, as shown in figure 8, the apparatus may include:
First Chinese word coding unit 10, the second Chinese word coding unit 20 and Vector Fusion unit 30 can further include map pretreatment
Unit 40.Wherein the major function of each component units is as follows:
First Chinese word coding unit 10 is responsible for carrying out Chinese word coding using attribute tags of the knowledge mapping to word each in sentence, obtains
First to each word expresses vector.
Specifically, the first Chinese word coding unit 10 may include: coupling subelement 11, participle subelement 12 and the first Chinese word coding
Subelement 13.
Wherein, coupling subelement 11 is responsible for utilizing the entity and the corresponding attribute of the entity in knowledge mapping identification sentence
Label.Specifically, coupling subelement 11 can be matched sentence using longest match principle in knowledge mapping, be identified
Entity in sentence.For example, each n-gram of available sentence, each value that wherein n is 1 or more.By each n-gram respectively with
Knowledge mapping is matched, and sees which n-gram is matched to entity in knowledge mapping, when there is the multiple n-gram to overlap
When being all matched to entity, take wherein the longest n-gram of length as the entity identified.
Participle subelement 12 is responsible for segmenting sentence using the recognition result of coupling subelement 11, and each to what is obtained
Word marks attribute tags.Subelement is segmented when segmenting to sentence, the entity that coupling subelement 11 can be identified
As independent word.
First Chinese word coding subelement 13 is responsible for carrying out Chinese word coding to the attribute tags of each word, such as can be to each word
Attribute tags carry out one-hot coding, and coding result is carried out to the conversion of full articulamentum, obtain the first of each word express to
Amount.
For the coding result of one-hot, the conversion of full articulamentum is carried out, it is therefore an objective to by the volume of the attribute tags of each word
Code result maps on entity tag, which is exactly the label that entity mark is carried out to word in sentence.By connecting entirely
After connecing layer conversion, the first expression vector of each word is obtained.
In embodiments of the present invention, above-mentioned full articulamentum can be trained in advance.Training process may include: preparatory
Using the sentence for having been marked with entity tag as training sample, above-mentioned reality is carried out to the sentence in training sample using knowledge mapping
After body identification, participle, the mark of attribute tags, one-hot coding, as the input of the full articulamentum, each word in the sentence
The first expression vector that corresponding entity tag is constituted is that the target of the full articulamentum exports, and is trained to full articulamentum.Instruction
The full articulamentum got is actually to carry out after one-hot coding coding result to the mapping of entity tag.
Map pretreatment unit 40 is responsible for integrating attribute tags of the entity each in knowledge mapping in each field, obtains
To the corresponding attribute tags set of each entity；The corresponding attribute tags set of each entity is stored in key assignments storage engines.Accordingly
Ground, coupling subelement 11 can be matched sentence using Longest prefix match algorithm in key assignments storage engines.
Second Chinese word coding unit 20 is responsible for carrying out Chinese word coding to word each in sentence based on sentence structure, obtains each word
Second expression vector.Specifically, the second Chinese word coding unit 20 can determine the term vector of each word in sentence first；Then by word
Vector input neural network trained in advance respectively obtains the second expression vector of each word.
Second Chinese word coding unit 20 when the term vector of each word, can be generated in determining sentence using existing term vector
Then tool, such as word2vec etc. can be directed to based on based on semantic training word2vec in advance using the word2vec
Each word generates term vector respectively, and the corresponding term vector length of each word is identical.The mode of this determining term vector is based on language
Justice, enable to the distance between term vector to embody the correlation degree between phrase semantic, correlation degree is higher between semanteme
Word, the distance between corresponding term vector are smaller.
Above-mentioned neural network can use such as two-way RNN (Recognition with Recurrent Neural Network), unidirectional RNN, CNN (convolutional Neural net
Network) etc..Wherein preferably bidirectional RNN.
Vector Fusion unit 30 is responsible for merging the first expression expression vector of vector sum second, obtains the reality to sentence
Body annotation results.
Specifically, the first expression vector sum second of each word can be expressed vector and carried out respectively by Vector Fusion unit 30
Splicing obtains the third expression vector of each word；Then the third expression vector of each word is converted to respectively by full articulamentum
The result vector of word, wherein the total quantity of the length correspondent entity label of result vector, every corresponding each reality of result vector
Body label, everybody value embody the score of correspondent entity label；Finally sentence is carried out according to the result vector of each word real
Body mark.
Wherein, Vector Fusion unit 30 is when the result vector according to each word carries out entity mark to sentence, Ke Yifen
Entity mark is not carried out to each word in sentence according to the entity tag of highest scoring in the result vector of each word.
Fig. 9 is intention assessment structure drawing of device provided in an embodiment of the present invention, as shown in figure 9, the apparatus may include: the
One coding unit, 50, second coding units 60 and Vector Fusion unit 70 can also include map pretreatment unit 80.Its
In each component units major function it is as follows:
First coding unit 50 is responsible for being combined coding to the attribute tags of word each in sentence using knowledge mapping,
Obtain first vector of sentence.
Wherein, first coding unit 50 can specifically include: coupling subelement 51, participle subelement 52 and assembly coding
Subelement 53.
Wherein, coupling subelement 51 is responsible for utilizing the entity and the corresponding attribute of the entity in knowledge mapping identification sentence
Label.Specifically, coupling subelement 51 can be matched sentence using Longest prefix match algorithm in knowledge mapping, be identified
Entity in sentence.
Participle subelement 52 is responsible for segmenting sentence using recognition result, and marks attribute mark to obtained each word
Label.Wherein, the entity identified coupling subelement 51 in participle is as independent word.
Assembly coding subelement 53 is responsible for being combined the attribute tags of each word coding, and coding result is carried out entirely
The conversion of articulamentum, obtains first vector of sentence, the total quantity of the length correspondent entity label of first vector, and first
Every value of vector is weight of the corresponding entity tag of this in sentence.
Map pretreatment unit 80 is responsible for integrating attribute tags of the entity each in knowledge mapping in each field, obtains
To the corresponding attribute tags set of each entity；The corresponding attribute tags set of each entity is stored in key assignments storage engines.Accordingly
Ground, above-mentioned coupling subelement 51 can be matched sentence using Longest prefix match algorithm in key assignments storage engines.
Second coding unit 60 is responsible for encoding sentence based on sentence structure, obtains second vector of sentence.
Specifically, second coding unit 60 can determine the term vector of each word in sentence first；Then term vector is inputted preparatory
Trained neural network obtains second vector of sentence.
Wherein in determining sentence when the term vector of each word, second coding unit 60 is using based on semantic training in advance
Word2vec, generate term vector respectively for word each in sentence.
Above-mentioned neural network can use such as two-way RNN (Recognition with Recurrent Neural Network), unidirectional RNN, CNN (convolutional Neural net
Network) etc..Wherein preferably bidirectional RNN.
Second coding unit 60 obtains second vector of sentence in the neural network that term vector input is trained in advance
When, the neural network that can specifically train term vector input in advance respectively obtains the second expression vector of each word；It will be last
Second vector of the second expression vector of one word as sentence.
Vector Fusion unit 70 is responsible for merging first vector sum, second vector of sentence, obtains to sentence
Intention assessment result.Specifically, first vector sum, second vector can be spliced, obtains third sentence vector；By
Three vectors are converted to result vector by full articulamentum, and wherein the length of result vector corresponds to the categorical measure of sentence intention,
Everybody corresponding all kinds of sentences of result vector are intended to, everybody value embodies the score that corresponding sentence is intended to；According to result vector
Determine that sentence is intended to.
Wherein, Vector Fusion unit 70, can be by score in result vector when determining that sentence is intended to according to result vector
Highest sentence is intended to be intended to as the sentence of sentence.
The method of above-mentioned entity mark and intention assessment can be applied to the several scenes based on natural language processing, herein
For the example of an application scenarios:
In intelligent answer field, such as " Zhou Jielun was drilled intelligent answer class client input problem of the user on mobile phone
Which film ", after above-mentioned entity mark and intention assessment, marking out entity " Zhou Jielun " is " Actor_name (performer's name
Word) ", it is intended that be " which film a performer drilled ".So corresponding processing logic of the intention is looked into movie database
Look for the corresponding movie name of entity that " Actor_name (actor names) " are labeled as in the sentence.Assuming that in movie database
Find " Zhou Jielun " corresponding movie name are as follows: " ineffable secret ", " thorn mound ", " balcony love ", " Gold world
First " ..., then intelligent answer class client directly can return to answer: " ineffable secret ", " thorn mound ", " day to user
Platform love ", " Curse of the Golden Flower " ....
Figure 10 schematically illustrates example apparatus 1000 according to various embodiments.Equipment 1000 may include one or more
A processor 1002, system control logic 1001 are coupled at least one processor 1002, nonvolatile memory (non-
Volatile memory, NMV)/memory 1004 is coupled in system control logic 1001, and network interface 1006 is coupled in system
Control logic 1001.
Processor 1002 may include one or more single core processors or multi-core processor.Processor 1002 may include any
The combination of general service processor or application specific processor (such as image processor, application processor baseband processor).
System control logic 1001 in one embodiment, it may include any interface controller appropriate, to provide everywhere
Any suitable interface of at least one of device 1002 is managed, and/or is provided any to being communicated with system control logic 1001
Any suitable interface of suitable equipment or component.
System control logic 1001 in one embodiment, it may include one or more Memory Controller Hub, to provide to being
The interface of system memory 1003.Installed System Memory 1003 is used to load and storing data and/or instruction.For example, corresponding equipment 1000,
In one embodiment, Installed System Memory 1003 may include any suitable volatile memory.
NVM/ memory 1004 may include the computer-readable medium of one or more tangible nonvolatiles, for storing number
According to and/or instruction.For example, NVM/ memory 1004 may include any suitable non-volatile memory device, it is such as one or more
Hard disk (hard disk device, HDD), one or more CDs (compact disk, CD), and/or one or more numbers
Word universal disc (digital versatile disk, DVD).
NVM/ memory 1004 may include storage resource, which is physically that the system is installed or can be with
A part of accessed equipment, but it is not necessarily a part of equipment.For example, NVM/ memory 1004 can be via network interface
1006 are accessed by network.
Installed System Memory 1003 and NVM/ memory 1004 can respectively include the copy of interim or lasting instruction 1010.
Instruction 1010 may include that equipment 1000 is caused to realize that Fig. 1 or Fig. 5 is described when being executed by least one of processor 1002
One of method or combined instruction.In each embodiment, instruction 1010 or hardware, firmware and/or component software can additionally/can
Alternatively it is placed in system control logic 1001, network interface 1006 and/or processor 1002.
Network interface 1006 may include a receiver to provide wireless interface and one or more nets for equipment 1000
Network and/or any suitable equipment are communicated.Network interface 1006 may include any suitable hardware and/or firmware.Network
Interface 1006 may include mutiple antennas to provide MIMO wireless interface.In one embodiment, network interface 1006
It may include a network adapter, a wireless network adapter, a telephone modem and/or wireless modulation-demodulation
Device.
In one embodiment, at least one of processor 1002 can be with one or more for system control logic
The logic of a controller encapsulates together.In one embodiment, at least one of processor can be patrolled with for system control
The logic for the one or more controllers collected is encapsulated together to form system in package.In one embodiment, in processor
At least one can be integrated on the same die with the logic of one or more controllers for system control logic.One
In a embodiment, at least one of processor can be with the logical set of one or more controllers for system control logic
At on the same die to form System on Chip/SoC.
Equipment 1000 can further comprise input/output device 1005.Input/output device 1005 may include user interface
It is intended to interact user with equipment 1000, it may include peripheral component interface is designed so that peripheral assembly can be with
System interaction, and/or, it may include sensor, it is intended to determine environmental condition and/or the location information in relation to equipment 1000.
In several embodiments provided by the present invention, it should be understood that disclosed system, device and method can be with
It realizes by another way.For example, the apparatus embodiments described above are merely exemplary, for example, the unit
It divides, only a kind of logical function partition, there may be another division manner in actual implementation.
The unit as illustrated by the separation member may or may not be physically separated, aobvious as unit
The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple
In network unit.It can select some or all of unit therein according to the actual needs to realize the mesh of this embodiment scheme
's.
It, can also be in addition, the functional units in various embodiments of the present invention may be integrated into one processing unit
It is that each unit physically exists alone, can also be integrated in one unit with two or more units.Above-mentioned integrated list
Member both can take the form of hardware realization, can also realize in the form of hardware adds SFU software functional unit.
The above-mentioned integrated unit being realized in the form of SFU software functional unit can store and computer-readable deposit at one
In storage media.Above-mentioned SFU software functional unit is stored in a storage medium, including some instructions are used so that a computer
It is each that equipment (can be personal computer, server or the network equipment etc.) or processor (processor) execute the present invention
The part steps of embodiment the method.And storage medium above-mentioned includes: USB flash disk, mobile hard disk, read-only memory (Read-
Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic or disk etc. it is various
It can store the medium of program code.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Within mind and principle, any modification, equivalent substitution, improvement and etc. done be should be included within the scope of the present invention.
Claims (46)
1. a kind of entity mask method, which is characterized in that this method comprises:
Using knowledge mapping at least partly the attribute tags of word carry out Chinese word coding in sentence, the of at least partly word is obtained
One expression vector；
Based on sentence structure at least partly word carries out Chinese word coding in the sentence, the second expression of at least partly word is obtained
Vector；
First expression vector sum second is expressed vector to merge, obtains the entity annotation results to the sentence.
2. the method according to claim 1, wherein it is described using knowledge mapping at least partly word in sentence
Attribute tags carry out Chinese word coding include:
The entity and the corresponding attribute tags of the entity in the sentence are identified using knowledge mapping；
The sentence is segmented using the recognition result, and attribute tags are marked to obtained at least partly word；
Chinese word coding is carried out to the attribute tags of at least partly word, and coding result is carried out to the conversion of full articulamentum, obtain to
First expression vector of small part word.
3. according to the method described in claim 2, it is characterized in that, identifying the entity packet in the sentence using knowledge mapping
It includes:
The sentence is matched in knowledge mapping using longest match principle, identifies the entity in the sentence.
4. according to the method described in claim 3, it is characterized in that, this method further include: by entity each in knowledge mapping each
Attribute tags in field are integrated, and the corresponding attribute tags set of each entity is obtained；By the corresponding attribute tags of each entity
Set is stored in key assignments storage engines；
It is described using Longest prefix match algorithm the sentence is carried out in knowledge mapping matching include: it is described using longest matching calculate
Method matches the sentence in the key assignments storage engines.
5. according to the method described in claim 2, it is characterized in that, carrying out participle packet to the sentence using the recognition result
It includes:
Sentence is segmented, wherein the entity that will identify that is as independent word.
6. according to the method described in claim 2, it is characterized in that, the attribute tags at least partly word carry out word volume
Code include:
Solely hot one-hot coding is carried out to the attribute tags of at least partly word.
7. the method according to claim 1, wherein it is described based on sentence structure in the sentence at least partly
Word carries out Chinese word coding
Determine in the sentence at least partly term vector of word；
By term vector input neural network trained in advance, the second expression vector of at least partly word is respectively obtained.
8. the method according to the description of claim 7 is characterized in that determining in the sentence at least partly term vector packet of word
It includes:
Using based on semantic word2vec trained in advance, term vector is generated respectively at least partly word in the sentence.
9. the method according to the description of claim 7 is characterized in that the neural network includes: bidirectional circulating neural network.
10. the method according to claim 1, wherein it is described by first expression vector sum second express vector into
Row fusion, obtain include: to the entity annotation results of the sentence
The first of at least partly word expression vector sum second is expressed vector to splice respectively, obtains at least partly word
Third expresses vector；
The third expression vector of at least partly word is converted to the result vector of at least partly word by full articulamentum,
In, the total quantity of the length correspondent entity label of the result vector, every corresponding each entity tag of the result vector, respectively
The value of position embodies the score of correspondent entity label；
Result vector according at least partly word carries out entity mark to the sentence.
11. according to the method described in claim 10, it is characterized in that, the result vector of the foundation at least partly word is to institute
Stating sentence progress entity mark includes:
Respectively according to the entity tag of highest scoring in the result vector of at least partly word at least partly word in sentence
Carry out entity mark.
12. a kind of intension recognizing method, which is characterized in that this method comprises:
It is combined coding using attribute tags of the knowledge mapping at least partly word in sentence, obtains the first of the sentence
Sentence vector；
The sentence is encoded based on sentence structure, obtains second vector of the sentence；
First vector sum, second vector of the sentence is merged, the intention assessment result to the sentence is obtained.
13. according to the method for claim 12, which is characterized in that it is described using knowledge mapping at least partly word in sentence
The attribute tags of language are combined coding
The entity and the corresponding attribute tags of the entity in the sentence are identified using knowledge mapping；
The sentence is segmented using the recognition result, and attribute tags are marked to obtained at least partly word；
Coding is combined to the attribute tags of at least partly word, and coding result is carried out to the conversion of full articulamentum, is obtained
First vector of the sentence, the total quantity of the length correspondent entity label of first vector, first vector
Every value be weight of the corresponding entity tag of this in the sentence.
14. according to the method for claim 13, which is characterized in that identify the entity packet in the sentence using knowledge mapping
It includes:
The sentence is matched in knowledge mapping using Longest prefix match algorithm, identifies the entity in the sentence.
15. according to the method for claim 13, which is characterized in that this method further include: entity each in knowledge mapping exists
Attribute tags in each field are integrated, and the corresponding attribute tags of each entity are obtained；The corresponding attribute tags of each entity are deposited
It is stored in key assignments storage engines；
It is described using Longest prefix match algorithm the sentence is carried out in knowledge mapping matching include: it is described using longest matching calculate
Method matches the sentence in the key assignments storage engines.
16. according to the method for claim 12, which is characterized in that described to be compiled based on sentence structure to the sentence
Code, second vector for obtaining the sentence include:
Determine in the sentence at least partly term vector of word；
By term vector input neural network trained in advance, second vector of the sentence is obtained.
17. according to the method for claim 16, which is characterized in that determine in the sentence at least partly term vector of word
Include:
Using based on semantic word2vec trained in advance, term vector is generated respectively at least partly word in the sentence.
18. according to the method for claim 16, which is characterized in that the neural network includes: bidirectional circulating neural network.
19. according to the method for claim 16, which is characterized in that by term vector input nerve net trained in advance
Network, second vector for obtaining the sentence include:
By term vector input neural network trained in advance, the second expression vector of at least partly word is respectively obtained；
Using the second expression vector of the last one word as second vector of the sentence.
20. according to the method for claim 12, which is characterized in that by first vector sum, second vector of the sentence
Merged, obtain include: to the intention assessment result of the sentence
Second vector described in first vector sum is spliced, third sentence vector is obtained；
The third sentence vector is converted into result vector by full articulamentum, wherein the length of the result vector corresponds to sentence
Everybody corresponding all kinds of sentences of the categorical measure of intention, the result vector are intended to, everybody value embodies corresponding sentence and is intended to
Score；
Determine that the sentence is intended to according to the result vector.
21. according to the method for claim 20, which is characterized in that determine that the sentence is intended to packet according to the result vector
It includes:
The sentence of highest scoring in the result vector is intended to be intended to as the sentence of the sentence.
22. a kind of entity annotation equipment, which is characterized in that the device includes:
First Chinese word coding unit, for using knowledge mapping in sentence at least partly word attribute tags carry out Chinese word coding,
Obtain the first expression vector of at least partly word；
Second Chinese word coding unit, for based on sentence structure in the sentence at least partly word carry out Chinese word coding, obtain to
Second expression vector of small part word；
Vector Fusion unit merges for the first expression vector sum second to be expressed vector, obtains the reality to the sentence
Body annotation results.
23. device according to claim 22, which is characterized in that the first Chinese word coding unit includes:
Coupling subelement, for identifying entity and the corresponding attribute tags of the entity in the sentence using knowledge mapping；
Subelement is segmented, for the recognition result using the coupling subelement sentence is segmented, and to obtaining
At least partly word marks attribute tags；
First Chinese word coding subelement carries out Chinese word coding for the attribute tags at least partly word, and coding result is carried out
The conversion of full articulamentum obtains the first expression vector of at least partly word.
24. device according to claim 23, which is characterized in that the coupling subelement is specifically used for:
The sentence is matched in knowledge mapping using longest match principle, identifies the entity in the sentence.
25. device according to claim 24, which is characterized in that the device further include:
Map pretreatment unit obtains each for integrating attribute tags of the entity each in knowledge mapping in each field
The corresponding attribute tags set of entity；The corresponding attribute tags set of each entity is stored in key assignments storage engines；
The coupling subelement is matched the sentence using Longest prefix match algorithm in the key assignments storage engines.
26. device according to claim 23, which is characterized in that the participle subelement is specifically used for: being carried out to sentence
Participle, wherein the entity that the coupling subelement is identified is as independent word.
27. device according to claim 23, which is characterized in that the first Chinese word coding subelement is at least partly word
It is specific to execute when attribute tags carry out Chinese word coding:
Solely hot one-hot coding is carried out to the attribute tags of at least partly word.
28. device according to claim 22, which is characterized in that the second Chinese word coding unit is specifically used for:
Determine in the sentence at least partly term vector of word；
By term vector input neural network trained in advance, the second expression vector of at least partly word is respectively obtained.
29. device according to claim 28, which is characterized in that the second Chinese word coding unit is in determining the sentence
At least partly the term vector of word when, it is specific to execute:
Using based on semantic word2vec trained in advance, term vector is generated respectively at least partly word in the sentence.
30. device according to claim 28, which is characterized in that the neural network includes: bidirectional circulating neural network.
31. device according to claim 22, which is characterized in that the Vector Fusion unit is specifically used for:
The first of at least partly word expression vector sum second is expressed vector to splice respectively, obtains at least partly word
Third expresses vector；
The third expression vector of at least partly word is converted to the result vector of at least partly word by full articulamentum,
In, the total quantity of the length correspondent entity label of the result vector, every corresponding each entity tag of the result vector, respectively
The value of position embodies the score of correspondent entity label；
Result vector according at least partly word carries out entity mark to the sentence.
32. device according to claim 31, which is characterized in that the Vector Fusion unit is according at least partly word
Result vector when carrying out entity mark to the sentence, it is specific to execute:
Respectively according to the entity tag of highest scoring in the result vector of at least partly word at least partly word in sentence
Carry out entity mark.
33. a kind of intention assessment device, which is characterized in that the device includes:
First coding unit, for being combined volume using attribute tags of the knowledge mapping at least partly word in sentence
Code, obtains first vector of the sentence；
Second coding unit, for being encoded to the sentence based on sentence structure, obtain second of the sentence to
Amount；
Vector Fusion unit is obtained for merging first vector sum, second vector of the sentence to the sentence
The intention assessment result of son.
34. device according to claim 33, which is characterized in that first coding unit specifically includes:
Coupling subelement, for identifying entity and the corresponding attribute tags of the entity in the sentence using knowledge mapping；
Subelement is segmented, for segmenting using the recognition result to the sentence, and to obtained at least partly word
Mark attribute tags；
Assembly coding subelement is combined coding for the attribute tags at least partly word, and coding result is carried out
The conversion of full articulamentum, obtains first vector of the sentence, the length correspondent entity label of first vector it is total
Quantity, every value of first vector are weight of the corresponding entity tag of this in the sentence.
35. device according to claim 34, which is characterized in that the coupling subelement is specifically used for utilizing knowledge graph
Entity in sentence described in spectrum discrimination includes:
The sentence is matched in knowledge mapping using Longest prefix match algorithm, identifies the entity in the sentence.
36. device according to claim 35, which is characterized in that the device further include:
Map pretreatment unit obtains each for integrating attribute tags of the entity each in knowledge mapping in each field
The corresponding attribute tags set of entity；The corresponding attribute tags set of each entity is stored in key assignments storage engines；
The coupling subelement is matched the sentence using Longest prefix match algorithm in the key assignments storage engines.
37. device according to claim 33, which is characterized in that second coding unit is specifically used for:
Determine in the sentence at least partly term vector of word；
By term vector input neural network trained in advance, second vector of the sentence is obtained.
38. the device according to claim 37, which is characterized in that second coding unit is in determining the sentence
At least partly the term vector of word when, it is specific to execute:
Using based on semantic word2vec trained in advance, term vector is generated respectively at least partly word in the sentence.
39. the device according to claim 38, which is characterized in that the neural network includes: bidirectional circulating neural network.
40. device according to claim 33, which is characterized in that second coding unit is specifically used for:
By term vector input neural network trained in advance, the second expression vector of at least partly word is respectively obtained；
Using the second expression vector of the last one word as second vector of the sentence.
41. device according to claim 33, which is characterized in that the Vector Fusion unit is specifically used for:
Second vector described in first vector sum is spliced, third sentence vector is obtained；
The third sentence vector is converted into result vector by full articulamentum, wherein the length of the result vector corresponds to sentence
Everybody corresponding all kinds of sentences of the categorical measure of intention, the result vector are intended to, everybody value embodies corresponding sentence and is intended to
Score；
Determine that the sentence is intended to according to the result vector.
42. device according to claim 41, which is characterized in that the Vector Fusion unit is according to the result vector
It is specific to execute when determining that the sentence is intended to:
The sentence of highest scoring in the result vector is intended to be intended to as the sentence of the sentence.
43. a kind of equipment, including
Memory, including one or more program；
One or more processor is coupled to the memory, executes one or more of programs, to realize such as right
It is required that the operation executed in 1 to 11 any claim the method.
44. a kind of equipment, including
Memory, including one or more program；
One or more processor is coupled to the memory, executes one or more of programs, to realize such as right
It is required that the operation executed in 12 to 21 any claim the methods.
45. a kind of computer storage medium, the computer storage medium is encoded with computer program, and described program is by one
When a or multiple computers execute, so that one or more of computers are executed as described in any claim of claim 1 to 11
The operation executed in method.
46. a kind of computer storage medium, the computer storage medium is encoded with computer program, and described program is by one
When a or multiple computers execute, so that one or more of computers are executed as described in any claim of claim 12 to 21
The operation executed in method.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710655187.2A CN109388793B (en) | 2017-08-03 | 2017-08-03 | Entity marking method, intention identification method, corresponding device and computer storage medium |
| PCT/CN2018/096640 WO2019024704A1 (en) | 2017-08-03 | 2018-07-23 | Entity annotation method, intention recognition method and corresponding devices, and computer storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710655187.2A CN109388793B (en) | 2017-08-03 | 2017-08-03 | Entity marking method, intention identification method, corresponding device and computer storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109388793A true CN109388793A (en) | 2019-02-26 |
| CN109388793B CN109388793B (en) | 2023-04-07 |
Family
ID=65233308
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710655187.2A Active CN109388793B (en) | 2017-08-03 | 2017-08-03 | Entity marking method, intention identification method, corresponding device and computer storage medium |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN109388793B (en) |
| WO (1) | WO2019024704A1 (en) |
Cited By (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109992671A (en) * | 2019-04-10 | 2019-07-09 | 出门问问信息科技有限公司 | Intension recognizing method, device, equipment and storage medium |
| CN110059320A (en) * | 2019-04-23 | 2019-07-26 | 腾讯科技（深圳）有限公司 | Entity relation extraction method, apparatus, computer equipment and storage medium |
| CN110188362A (en) * | 2019-06-10 | 2019-08-30 | 北京百度网讯科技有限公司 | Text handling method and device |
| CN110377691A (en) * | 2019-07-23 | 2019-10-25 | 上海应用技术大学 | Method, apparatus, equipment and the storage medium of text classification |
| CN110472230A (en) * | 2019-07-11 | 2019-11-19 | 平安科技（深圳）有限公司 | The recognition methods of Chinese text and device |
| CN110738041A (en) * | 2019-10-16 | 2020-01-31 | 天津市爱贝叶斯信息技术有限公司 | statement labeling method, device, server and storage medium |
| CN111027667A (en) * | 2019-12-06 | 2020-04-17 | 北京金山安全软件有限公司 | Intention category identification method and device |
| CN111104803A (en) * | 2019-12-31 | 2020-05-05 | 科大讯飞股份有限公司 | Semantic understanding processing method, device and equipment and readable storage medium |
| CN111159546A (en) * | 2019-12-24 | 2020-05-15 | 腾讯科技（深圳）有限公司 | Event pushing method and device, computer readable storage medium and computer equipment |
| CN111274815A (en) * | 2020-01-15 | 2020-06-12 | 北京百度网讯科技有限公司 | Method and device for mining entity attention points in text |
| CN111309872A (en) * | 2020-03-26 | 2020-06-19 | 北京百度网讯科技有限公司 | Search processing method, device and equipment |
| CN111353310A (en) * | 2020-02-28 | 2020-06-30 | 腾讯科技（深圳）有限公司 | Named entity identification method and device based on artificial intelligence and electronic equipment |
| CN111613341A (en) * | 2020-05-22 | 2020-09-01 | 云知声智能科技股份有限公司 | Entity linking method and device based on semantic components |
| CN111753024A (en) * | 2020-06-24 | 2020-10-09 | 河北工程大学 | Public safety field-oriented multi-source heterogeneous data entity alignment method |
| CN111813447A (en) * | 2019-04-12 | 2020-10-23 | 杭州中天微系统有限公司 | Processing method and processing device for data splicing instruction |
| CN112201250A (en) * | 2020-09-30 | 2021-01-08 | 中移（杭州）信息技术有限公司 | Semantic analysis method and device, electronic equipment and storage medium |
| CN112364664A (en) * | 2020-11-19 | 2021-02-12 | 北京京东尚科信息技术有限公司 | Method and device for training intention recognition model and intention recognition and storage medium |
| CN112543932A (en) * | 2020-01-22 | 2021-03-23 | 华为技术有限公司 | Semantic analysis method, device, equipment and storage medium |
| CN112749556A (en) * | 2020-08-04 | 2021-05-04 | 腾讯科技（深圳）有限公司 | Multi-language model training method and device, storage medium and electronic equipment |
| CN113064997A (en) * | 2021-04-22 | 2021-07-02 | 中国平安财产保险股份有限公司 | Intent analysis method, device, equipment and medium based on BERT model |
| CN113157892A (en) * | 2021-05-24 | 2021-07-23 | 中国平安人寿保险股份有限公司 | User intention processing method and device, computer equipment and storage medium |
| CN113343692A (en) * | 2021-07-15 | 2021-09-03 | 杭州网易云音乐科技有限公司 | Search intention recognition method, model training method, device, medium and equipment |
| CN113360751A (en) * | 2020-03-06 | 2021-09-07 | 百度在线网络技术（北京）有限公司 | Intention recognition method, apparatus, device and medium |
| CN113377969A (en) * | 2021-08-16 | 2021-09-10 | 中航信移动科技有限公司 | Intention recognition data processing system |
| CN113505587A (en) * | 2021-06-23 | 2021-10-15 | 科大讯飞华南人工智能研究院（广州）有限公司 | Entity extraction method, related device, equipment and storage medium |
| CN113705236A (en) * | 2021-04-02 | 2021-11-26 | 腾讯科技（深圳）有限公司 | Entity comparison method, device, equipment and computer readable storage medium |
| CN113723114A (en) * | 2021-08-31 | 2021-11-30 | 平安普惠企业管理有限公司 | Semantic analysis method, device and equipment based on multi-intent recognition and storage medium |
Families Citing this family (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109840280B (en) * | 2019-03-05 | 2023-07-18 | 百度在线网络技术（北京）有限公司 | Text classification method and device and computer readable storage medium |
| CN111859976A (en) * | 2019-04-30 | 2020-10-30 | 广东小天才科技有限公司 | Method and device for expanding regular expression based on knowledge graph |
| CN110245198B (en) * | 2019-06-18 | 2021-08-27 | 北京百度网讯科技有限公司 | Multi-source ticketing data management method and system, server and computer readable medium |
| CN110597961B (en) * | 2019-09-18 | 2023-10-27 | 腾讯云计算（北京）有限责任公司 | Text category labeling method and device, electronic equipment and storage medium |
| CN114175017A (en) * | 2019-10-30 | 2022-03-11 | 深圳市欢太科技有限公司 | Model construction method, classification method, device, storage medium and electronic equipment |
| CN111753495A (en) * | 2019-11-07 | 2020-10-09 | 北京沃东天骏信息技术有限公司 | Method, device, equipment and storage medium for constructing prediction model of intention statement |
| CN111160033B (en) * | 2019-12-18 | 2024-02-27 | 车智互联（北京）科技有限公司 | Named entity identification method based on neural network, computing equipment and storage medium |
| CN111124350B (en) * | 2019-12-20 | 2023-10-27 | 科大讯飞股份有限公司 | Skill determination method and related equipment |
| CN111160034B (en) * | 2019-12-31 | 2024-02-27 | 东软集团股份有限公司 | Entity word labeling method, device, storage medium and equipment |
| CN111368527B (en) * | 2020-02-28 | 2023-06-20 | 上海汇航捷讯网络科技有限公司 | Key value matching method |
| CN111680207B (en) * | 2020-03-11 | 2023-08-04 | 华中科技大学鄂州工业技术研究院 | Method and device for determining search intention of user |
| US11379666B2 (en) | 2020-04-08 | 2022-07-05 | International Business Machines Corporation | Suggestion of new entity types with discriminative term importance analysis |
| CN111400480B (en) * | 2020-04-21 | 2023-05-12 | 支付宝(杭州)信息技术有限公司 | User intention recognition method and device for multi-round dialogue |
| CN113642302B (en) * | 2020-04-27 | 2024-04-02 | 阿里巴巴集团控股有限公司 | Training method and device for text filling model, text processing method and device |
| CN113742523B (en) * | 2020-05-29 | 2023-06-27 | 北京百度网讯科技有限公司 | Labeling method and device for text core entity |
| CN111695345B (en) * | 2020-06-12 | 2024-02-23 | 腾讯科技（深圳）有限公司 | Method and device for identifying entity in text |
| CN111708873B (en) * | 2020-06-15 | 2023-11-24 | 腾讯科技（深圳）有限公司 | Intelligent question-answering method, intelligent question-answering device, computer equipment and storage medium |
| CN111767726B (en) * | 2020-06-24 | 2024-02-06 | 北京奇艺世纪科技有限公司 | Data processing method and device |
| CN111832282B (en) * | 2020-07-16 | 2023-04-14 | 平安科技（深圳）有限公司 | External knowledge fused BERT model fine adjustment method and device and computer equipment |
| CN111797245B (en) * | 2020-07-27 | 2023-07-25 | 中国平安人寿保险股份有限公司 | Knowledge graph model-based information matching method and related device |
| CN111914568B (en) * | 2020-07-31 | 2024-02-06 | 平安科技（深圳）有限公司 | Method, device and equipment for generating text sentence and readable storage medium |
| CN111950288B (en) * | 2020-08-25 | 2024-02-23 | 海信视像科技股份有限公司 | Entity labeling method in named entity recognition and intelligent device |
| CN112100397A (en) * | 2020-09-07 | 2020-12-18 | 南京航空航天大学 | Electric power plan knowledge graph construction method and system based on bidirectional gating circulation unit |
| CN112015921B (en) * | 2020-09-15 | 2024-04-16 | 重庆广播电视大学重庆工商职业学院 | Natural language processing method based on learning auxiliary knowledge graph |
| CN112101009B (en) * | 2020-09-23 | 2024-03-26 | 中国农业大学 | Method for judging similarity of red-building dream character relationship frames based on knowledge graph |
| CN112699685B (en) * | 2021-01-08 | 2024-03-29 | 北京工业大学 | Named entity recognition method based on label-guided word fusion |
| CN113378574B (en) * | 2021-06-30 | 2023-10-24 | 武汉大学 | KGANN-based named entity identification method |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105095195A (en) * | 2015-07-03 | 2015-11-25 | 北京京东尚科信息技术有限公司 | Method and system for human-machine questioning and answering based on knowledge graph |
| CN105117487A (en) * | 2015-09-19 | 2015-12-02 | 杭州电子科技大学 | Book semantic retrieval method based on content structures |
| CN105335519A (en) * | 2015-11-18 | 2016-02-17 | 百度在线网络技术（北京）有限公司 | Model generation method and device as well as recommendation method and device |
| US20160132648A1 (en) * | 2014-11-06 | 2016-05-12 | ezDI, LLC | Data Processing System and Method for Computer-Assisted Coding of Natural Language Medical Text |
| US9367608B1 (en) * | 2009-01-07 | 2016-06-14 | Guangsheng Zhang | System and methods for searching objects and providing answers to queries using association data |
| CN106649394A (en) * | 2015-11-03 | 2017-05-10 | 中兴通讯股份有限公司 | Fusion knowledge base processing method and device and knowledge base management system |
| CN106776711A (en) * | 2016-11-14 | 2017-05-31 | 浙江大学 | A kind of Chinese medical knowledge mapping construction method based on deep learning |
| CN106776562A (en) * | 2016-12-20 | 2017-05-31 | 上海智臻智能网络科技股份有限公司 | A kind of keyword extracting method and extraction system |
| CN106815252A (en) * | 2015-12-01 | 2017-06-09 | 阿里巴巴集团控股有限公司 | A kind of searching method and equipment |
| CN106815192A (en) * | 2015-11-27 | 2017-06-09 | 北京国双科技有限公司 | Model training method and device and sentence emotion identification method and device |
| CN106875940A (en) * | 2017-03-06 | 2017-06-20 | 吉林省盛创科技有限公司 | A kind of Machine self-learning based on neutral net builds knowledge mapping training method |
| CN106897568A (en) * | 2017-02-28 | 2017-06-27 | 北京大数医达科技有限公司 | The treating method and apparatus of case history structuring |
-
2017
- 2017-08-03 CN CN201710655187.2A patent/CN109388793B/en active Active
-
2018
- 2018-07-23 WO PCT/CN2018/096640 patent/WO2019024704A1/en active Application Filing
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9367608B1 (en) * | 2009-01-07 | 2016-06-14 | Guangsheng Zhang | System and methods for searching objects and providing answers to queries using association data |
| US20160132648A1 (en) * | 2014-11-06 | 2016-05-12 | ezDI, LLC | Data Processing System and Method for Computer-Assisted Coding of Natural Language Medical Text |
| CN105095195A (en) * | 2015-07-03 | 2015-11-25 | 北京京东尚科信息技术有限公司 | Method and system for human-machine questioning and answering based on knowledge graph |
| CN105117487A (en) * | 2015-09-19 | 2015-12-02 | 杭州电子科技大学 | Book semantic retrieval method based on content structures |
| WO2017076263A1 (en) * | 2015-11-03 | 2017-05-11 | 中兴通讯股份有限公司 | Method and device for integrating knowledge bases, knowledge base management system and storage medium |
| CN106649394A (en) * | 2015-11-03 | 2017-05-10 | 中兴通讯股份有限公司 | Fusion knowledge base processing method and device and knowledge base management system |
| CN105335519A (en) * | 2015-11-18 | 2016-02-17 | 百度在线网络技术（北京）有限公司 | Model generation method and device as well as recommendation method and device |
| CN106815192A (en) * | 2015-11-27 | 2017-06-09 | 北京国双科技有限公司 | Model training method and device and sentence emotion identification method and device |
| CN106815252A (en) * | 2015-12-01 | 2017-06-09 | 阿里巴巴集团控股有限公司 | A kind of searching method and equipment |
| CN106776711A (en) * | 2016-11-14 | 2017-05-31 | 浙江大学 | A kind of Chinese medical knowledge mapping construction method based on deep learning |
| CN106776562A (en) * | 2016-12-20 | 2017-05-31 | 上海智臻智能网络科技股份有限公司 | A kind of keyword extracting method and extraction system |
| CN106897568A (en) * | 2017-02-28 | 2017-06-27 | 北京大数医达科技有限公司 | The treating method and apparatus of case history structuring |
| CN106875940A (en) * | 2017-03-06 | 2017-06-20 | 吉林省盛创科技有限公司 | A kind of Machine self-learning based on neutral net builds knowledge mapping training method |
Non-Patent Citations (2)
| Title |
|---|
| 刘康等: "基于表示学习的知识库问答研究进展与展望", 《自动化学报》 * |
| 刘玉娇等: "基于深度学习的中文微博命名实体识别", 《四川大学学报(工程科学版)》 * |
Cited By (45)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109992671A (en) * | 2019-04-10 | 2019-07-09 | 出门问问信息科技有限公司 | Intension recognizing method, device, equipment and storage medium |
| CN111813447B (en) * | 2019-04-12 | 2022-11-08 | 杭州中天微系统有限公司 | Processing method and processing device for data splicing instruction |
| CN111813447A (en) * | 2019-04-12 | 2020-10-23 | 杭州中天微系统有限公司 | Processing method and processing device for data splicing instruction |
| CN110059320A (en) * | 2019-04-23 | 2019-07-26 | 腾讯科技（深圳）有限公司 | Entity relation extraction method, apparatus, computer equipment and storage medium |
| CN110188362A (en) * | 2019-06-10 | 2019-08-30 | 北京百度网讯科技有限公司 | Text handling method and device |
| CN110188362B (en) * | 2019-06-10 | 2021-04-20 | 北京百度网讯科技有限公司 | Text processing method and device |
| CN110472230B (en) * | 2019-07-11 | 2023-09-05 | 平安科技（深圳）有限公司 | Chinese text recognition method and device |
| CN110472230A (en) * | 2019-07-11 | 2019-11-19 | 平安科技（深圳）有限公司 | The recognition methods of Chinese text and device |
| CN110377691A (en) * | 2019-07-23 | 2019-10-25 | 上海应用技术大学 | Method, apparatus, equipment and the storage medium of text classification |
| CN110738041B (en) * | 2019-10-16 | 2023-12-01 | 天津市爱贝叶斯信息技术有限公司 | Statement labeling method, device, server and storage medium |
| CN110738041A (en) * | 2019-10-16 | 2020-01-31 | 天津市爱贝叶斯信息技术有限公司 | statement labeling method, device, server and storage medium |
| CN111027667A (en) * | 2019-12-06 | 2020-04-17 | 北京金山安全软件有限公司 | Intention category identification method and device |
| CN111027667B (en) * | 2019-12-06 | 2023-10-17 | 北京金山安全软件有限公司 | Method and device for identifying intention category |
| CN111159546B (en) * | 2019-12-24 | 2023-10-24 | 深圳市雅阅科技有限公司 | Event pushing method, event pushing device, computer readable storage medium and computer equipment |
| CN111159546A (en) * | 2019-12-24 | 2020-05-15 | 腾讯科技（深圳）有限公司 | Event pushing method and device, computer readable storage medium and computer equipment |
| CN111104803B (en) * | 2019-12-31 | 2024-02-13 | 科大讯飞股份有限公司 | Semantic understanding processing method, device, equipment and readable storage medium |
| CN111104803A (en) * | 2019-12-31 | 2020-05-05 | 科大讯飞股份有限公司 | Semantic understanding processing method, device and equipment and readable storage medium |
| CN111274815A (en) * | 2020-01-15 | 2020-06-12 | 北京百度网讯科技有限公司 | Method and device for mining entity attention points in text |
| CN111274815B (en) * | 2020-01-15 | 2024-04-12 | 北京百度网讯科技有限公司 | Method and device for mining entity focus point in text |
| US11775761B2 (en) | 2020-01-15 | 2023-10-03 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Method and apparatus for mining entity focus in text |
| CN112543932A (en) * | 2020-01-22 | 2021-03-23 | 华为技术有限公司 | Semantic analysis method, device, equipment and storage medium |
| CN111353310A (en) * | 2020-02-28 | 2020-06-30 | 腾讯科技（深圳）有限公司 | Named entity identification method and device based on artificial intelligence and electronic equipment |
| CN111353310B (en) * | 2020-02-28 | 2023-08-11 | 腾讯科技（深圳）有限公司 | Named entity identification method and device based on artificial intelligence and electronic equipment |
| CN113360751A (en) * | 2020-03-06 | 2021-09-07 | 百度在线网络技术（北京）有限公司 | Intention recognition method, apparatus, device and medium |
| CN111309872A (en) * | 2020-03-26 | 2020-06-19 | 北京百度网讯科技有限公司 | Search processing method, device and equipment |
| CN111309872B (en) * | 2020-03-26 | 2023-08-08 | 北京百度网讯科技有限公司 | Search processing method, device and equipment |
| CN111613341A (en) * | 2020-05-22 | 2020-09-01 | 云知声智能科技股份有限公司 | Entity linking method and device based on semantic components |
| CN111613341B (en) * | 2020-05-22 | 2024-02-02 | 云知声智能科技股份有限公司 | Entity linking method and device based on semantic components |
| CN111753024B (en) * | 2020-06-24 | 2024-02-20 | 河北工程大学 | Multi-source heterogeneous data entity alignment method oriented to public safety field |
| CN111753024A (en) * | 2020-06-24 | 2020-10-09 | 河北工程大学 | Public safety field-oriented multi-source heterogeneous data entity alignment method |
| CN112749556B (en) * | 2020-08-04 | 2022-09-13 | 腾讯科技（深圳）有限公司 | Multi-language model training method and device, storage medium and electronic equipment |
| CN112749556A (en) * | 2020-08-04 | 2021-05-04 | 腾讯科技（深圳）有限公司 | Multi-language model training method and device, storage medium and electronic equipment |
| CN112201250A (en) * | 2020-09-30 | 2021-01-08 | 中移（杭州）信息技术有限公司 | Semantic analysis method and device, electronic equipment and storage medium |
| CN112201250B (en) * | 2020-09-30 | 2024-03-19 | 中移（杭州）信息技术有限公司 | Semantic analysis method and device, electronic equipment and storage medium |
| CN112364664A (en) * | 2020-11-19 | 2021-02-12 | 北京京东尚科信息技术有限公司 | Method and device for training intention recognition model and intention recognition and storage medium |
| CN112364664B (en) * | 2020-11-19 | 2023-12-05 | 北京京东尚科信息技术有限公司 | Training of intention recognition model, intention recognition method, device and storage medium |
| CN113705236A (en) * | 2021-04-02 | 2021-11-26 | 腾讯科技（深圳）有限公司 | Entity comparison method, device, equipment and computer readable storage medium |
| CN113064997A (en) * | 2021-04-22 | 2021-07-02 | 中国平安财产保险股份有限公司 | Intent analysis method, device, equipment and medium based on BERT model |
| CN113157892A (en) * | 2021-05-24 | 2021-07-23 | 中国平安人寿保险股份有限公司 | User intention processing method and device, computer equipment and storage medium |
| CN113505587A (en) * | 2021-06-23 | 2021-10-15 | 科大讯飞华南人工智能研究院（广州）有限公司 | Entity extraction method, related device, equipment and storage medium |
| CN113505587B (en) * | 2021-06-23 | 2024-04-09 | 科大讯飞华南人工智能研究院（广州）有限公司 | Entity extraction method, related device, equipment and storage medium |
| CN113343692A (en) * | 2021-07-15 | 2021-09-03 | 杭州网易云音乐科技有限公司 | Search intention recognition method, model training method, device, medium and equipment |
| CN113343692B (en) * | 2021-07-15 | 2023-09-12 | 杭州网易云音乐科技有限公司 | Search intention recognition method, model training method, device, medium and equipment |
| CN113377969A (en) * | 2021-08-16 | 2021-09-10 | 中航信移动科技有限公司 | Intention recognition data processing system |
| CN113723114A (en) * | 2021-08-31 | 2021-11-30 | 平安普惠企业管理有限公司 | Semantic analysis method, device and equipment based on multi-intent recognition and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109388793B (en) | 2023-04-07 |
| WO2019024704A1 (en) | 2019-02-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109388793A (en) | Entity mask method, intension recognizing method and corresponding intrument, computer storage medium | |
| US11853879B2 (en) | Generating vector representations of documents | |
| US11514247B2 (en) | Method, apparatus, computer device and readable medium for knowledge hierarchical extraction of a text | |
| US10437929B2 (en) | Method and system for processing an input query using a forward and a backward neural network specific to unigrams | |
| CN109522553A (en) | Name recognition methods and the device of entity | |
| CN113377971B (en) | Multimedia resource generation method and device, electronic equipment and storage medium | |
| CN105869642A (en) | Voice text error correction method and device | |
| US10803380B2 (en) | Generating vector representations of documents | |
| Siddique et al. | Linguistically-enriched and context-awarezero-shot slot filling | |
| CN109241319A (en) | A kind of picture retrieval method, device, server and storage medium | |
| CN109214417A (en) | The method for digging and device, computer equipment and readable medium that user is intended to | |
| CN113051914A (en) | Enterprise hidden label extraction method and device based on multi-feature dynamic portrait | |
| CN113836992B (en) | Label identification method, label identification model training method, device and equipment | |
| CN109271624A (en) | A kind of target word determines method, apparatus and storage medium | |
| CN113343692B (en) | Search intention recognition method, model training method, device, medium and equipment | |
| WO2021159812A1 (en) | Cancer staging information processing method and apparatus, and storage medium | |
| CN109660621A (en) | A kind of content delivery method and service equipment | |
| CN113761188A (en) | Text label determination method and device, computer equipment and storage medium | |
| CN112784156A (en) | Search feedback method, system, device and storage medium based on intention recognition | |
| CN110874408B (en) | Model training method, text recognition device and computing equipment | |
| US20230153533A1 (en) | Pre-training techniques for entity extraction in low resource domains | |
| CN116955707A (en) | Content tag determination method, device, equipment, medium and program product | |
| Amiriparian et al. | Humans inside: cooperative big multimedia data mining | |
| CN114329064A (en) | Video processing method, video processing device, computer equipment and storage medium | |
| CN106570116A (en) | Aggregation method and device for search results based on artificial intelligence |
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 |