CN104142917A - Hierarchical semantic tree construction method and system for language understanding - Google Patents
Hierarchical semantic tree construction method and system for language understanding Download PDFInfo
- Publication number
- CN104142917A CN104142917A CN201410216929.8A CN201410216929A CN104142917A CN 104142917 A CN104142917 A CN 104142917A CN 201410216929 A CN201410216929 A CN 201410216929A CN 104142917 A CN104142917 A CN 104142917A
- Authority
- CN
- China
- Prior art keywords
- semantic
- word
- node
- level
- statement
- 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
Abstract
The invention discloses a hierarchical semantic tree construction method and system for language understanding. The method mainly comprises the steps as follows: segmenting terms of a statement and loading a semantic knowledge base; recognizing all nodes of the statement according to an LV rule, and recognizing the level of the nodes according to semantic knowledge and term positions and collocations; generating a special node by punctuation at the end of the statement, and taking the special node as a root node of a semantic tree; merging the nodes according to generated node information, recognizing semantic side chunks of the statement, and taking a level-0 semantic side as a child node to be hung on the root node; circularly traversing all child nodes of the statement till no low-level semantic side exists, and taking the child nodes as leaf nodes to be hung on the child node. According to the hierarchical semantic tree construction method and system, under the condition that no syntactic resource exists, the semantic structure tree is obtained through semantic information and the term positions and collocations only, so that a computer can enter a deep semantic layer of a natural language, various processing of the natural language can be finished on the basis of understanding, the first step of semantic understanding of the natural language is realized, and the hierarchical semantic tree construction method and system can be applied to information retrieval, automatic abstraction, machine translation, text categorization, information filtration and the like.
Description
Technical field
The present invention relates to a kind of natural language processing field, specifically utilize the position of semantic knowledge and word and collocation and the Layer semantics tree constructing method and the system that obtain.
Background technology
Along with the development of electronic information technology, digital information resources is more and more widely used.This just needs machine also can understand natural language, completes the various processing to natural language, as information retrieval, automatic abstract, mechanical translation, text classification and information filtering etc. on the basis of " understanding ".Visible, to make computing machine can enter natural language semantic deep layer is the condition that achieves the above object.Want to allow machine understand the meaning of natural language, first will understand the structure of natural language statement, sentence structure is a kind of basic structure of natural language, generally comprises syntactic structure and semantic structure.In order better the semanteme of statement to be described, adopting sentence structure tree is a kind of simple and clear effective mode.The structure tree type of statement mainly comprises two kinds: one is syntactic structure tree, and one is semantic structure tree.Syntactic structure tree mainly comprises phrase structure tree, dependency tree etc., and it builds mainly automatically on basis of syntax mark, adopts the method based on statistics to realize, and the structure of this type of syntactic structure tree does not use or the semantic knowledge of less use word.
The structure of semantic structure tree must use semantic knowledge, building semantic tree is under the guidance of HNC (hierarchical network of concepts) theory, in the situation that there is no syntax resource, only use semantic knowledge and words and phrases language position and collocation to carry out, make computing machine can enter the semantic deep layer of natural language, on the basis of understanding, carry out the various processing of natural language, realize the first step of semantics comprehension on natural language, for subsequent applications creates conditions in the processes such as information retrieval, mechanical translation, information filtering, text classification.
A kind of method and apparatus from Text Flag semantic structure is disclosed in Chinese patent literature CN1606004A, form at least two candidate's semantic structures, likelihood based on described semantic structure is determined semantic score to each candidate's semantic structure, also based on word, syntax score is determined to each semantic structure in the position in text and the semantic entity that forms from this word position this semantic structure, syntax score and semantic score is combined at least a portion of the text is selected to semantic structure.In this scheme, define the pattern of entity, this pattern comprises semantic type and probability, Markov method and semantic rules, obtaining of these semantic contents need to be trained large-scale data, field dependence to text is strong, due to the complicacy of task, the effect obtaining is not necessarily desirable, and follow-up all operations all rely on the result of this step, and its effect will be had a greatly reduced quality.
Summary of the invention
Technical matters to be solved by this invention is that the method for mark semantic structure of the prior art need to train large-scale data, strong to the field dependence of text, thereby proposes a kind of Layer semantics tree constructing method and system without training.
For solving the problems of the technologies described above, the invention provides a kind of Layer semantics tree constructing method and system for language understanding, comprise the steps:
S1, input pending statement, pending statement is carried out to participle, and load the semantic knowledge of word after participle;
S2, according to word segmentation result, identify the semantic node of this statement;
S3, utilize semantic knowledge and word position and collocation to obtain the level of semantic node;
S4, identify the semantic limit of different levels in this statement;
S5, generate level semantic tree according to the semantic limit of each level.
Preferably, in described step S1, when pending statement is carried out to participle, according to field dictionary and universaling dictionary, pending statement is carried out to participle.
Preferably, described semantic knowledge comprises generalized concept class and the subclass thereof of word, and the generalized concept class of described word comprises dynamically, static state, thing, people, attribute, logic.
Preferably, the process that " according to word segmentation result, identifies the semantic node of this statement " in described step S2, comprising:
For the word after participle, if having logical concept in the semantic knowledge of word, this word is labeled as to L, if having dynamic concept in the semantic knowledge of word, be labeled as V;
To all words that are labeled as L or V, carry out LV and get rid of processing;
All L marks are carried out to mark according to its concept classification, and judge whether it has rear mark, if there is rear mark, the word of rear mark is labeled as to L1H, according to above-mentioned all mark generative semantics nodes.
Preferably, the process that " according to word segmentation result, identifies the semantic node of this statement " in described step S2, also comprises: using end of the sentence punctuate generative semantics node as root node.
Preferably, the process of " utilizing semantic knowledge and word position and collocation to obtain the level of semantic node " in described step S3, comprising:
The acquiescence level of all L marks and v mark is all designated as 0, and in the time occurring that two above-mentioned marks are adjacent, it is-1 that the level of second mark reduces one deck.
Preferably, the process of " identifying the semantic limit of different levels in this statement " in described step S4, comprises
To all semantic nodes that are labeled as V, carry out the identification of core verb, generate language piece;
To all semantic nodes that are labeled as L, generate language piece;
According to language piece generative semantics limit.
The process of carrying out the identification of core verb preferably, comprises:
Eliminating can not form the word of core verb;
Remaining word is given different weights according to the feature forming and word itself has, and selects core verb according to the ranking results of weights and positional information.
Preferably, the described process that generates level semantic tree according to the semantic limit of each level, comprising:
Select root node;
Language piece high level, according to the order in this level, be suspended on root node, as child node;
Travel through all child nodes, all language pieces within the scope of each child node are as the child node of this child node, until do not have new child node to produce.
The Layer semantics tree constructing system that described Layer semantics tree constructing method is corresponding, comprising:
Pretreatment unit: input pending statement, pending statement is carried out to participle, and load the semantic knowledge of word after participle;
First ray generation unit: according to word segmentation result, identify the semantic node of this statement; Utilize semantic knowledge and word position and collocation to obtain the level of semantic node;
The second sequence generating unit: the semantic limit of identifying different levels in this statement;
Layer semantics tree generation unit: generate level semantic tree according to the semantic limit of each level.
Technique scheme of the present invention has the following advantages compared to existing technology,
(1) the Layer semantics tree constructing method described in the present embodiment, mainly comprises the process that pre-service, node recognition, the identification of semantic limit, semantic tree generate, and output said structure tree can obtain Layer semantics tree.The scheme that Layer semantics tree in the present embodiment builds, all utilizes rule and method to realize to the analysis of statement.In this programme, by the identification of node and the identification of level, semantic limit and level thereof, carry out the scheduling of control law in different levels, different phase.Under this principle instructs, first need rule to carry out hierarchical classification, each rule-like only calls in the fixing level of analysis, and each rule is only paid close attention to the analysis to language phenomenon in contiguous language string, do not need to take into account the judgement to overall situation, but solve regular compatibility issue by scheduling.
(2) the Layer semantics tree constructing method in the present invention, in the situation that there is no syntax resource, the semantic structure tree that only uses semantic information and word position and collocation and obtain, make computing machine can enter the semantic deep layer of natural language, on the basis of understanding, complete the various processing to natural language, realized the first step of semantics comprehension on natural language.Build semantic tree, can be widely used in natural language processing field, as convenient in information retrieval, automatic abstract, mechanical translation, text classification and information filtering etc.The construction method of the semantic tree in the present embodiment, has been applied in patent documentation Chinese-English machine translation, has significantly improved readability and the accuracy of patent documentation translation.
Brief description of the drawings
For content of the present invention is more likely to be clearly understood, below according to a particular embodiment of the invention and by reference to the accompanying drawings, the present invention is further detailed explanation, wherein
Fig. 1 is the process flow diagram of Layer semantics tree constructing method of the present invention;
The node product process figure of Fig. 2 Layer semantics tree constructing method of the present invention;
The semantic limit product process figure of Fig. 3 Layer semantics tree constructing method of the present invention;
The result schematic diagram of an application example of Fig. 4, Fig. 5 Layer semantics tree constructing method of the present invention;
Fig. 6 is the structured flowchart of Layer semantics tree constructing system of the present invention.
Embodiment
embodiment 1:
A kind of Layer semantics tree constructing method and system for language understanding is provided in the present embodiment, semantic tree is semantic structure tree, be for a sentence in natural language, refer to feature language piece in a sentence (core verb language piece) and by the semantic relation between other language pieces of its decision.If the feature language piece V in a sentence is the verb that represents effect, this feature language piece determines must have actor language piece, object language piece, content language piece in this sentence, and sentence is semantic just complete only so.Although rear three can omit one in certain context environmental, these four kinds of language pieces are that sentence becomes the necessary member of complete semantic immediately, are again main language piece.And compare, auxiliary language piece is not the necessary member that sentence is set up, and is mainly the mode, means, approach, condition, time etc. that represent action.Main language piece and auxiliary language piece all can be pointed out by certain logical concept, and the semantic structure that therefore uses LV (logical concept and dynamic concept) criterion to identify sentence becomes possibility.Layer semantics tree constructing method in the present embodiment, utilize exactly LV criterion to identify main language piece and the auxiliary language piece of a sentence, this scheme can realize automatically statement is divided, and for Language Translation, can greatly improve readability and the accuracy of mechanical translation.
Layer semantics tree constructing method in the present embodiment, main processing procedure comprises: pending statement S110 obtains semantic tree S160 through pre-service S120, node recognition S130, semantic limit identification S140, semantic tree after generating S150, process flow diagram as shown in Figure 1, specifically comprises the steps:
S1, input pending statement, pending statement is carried out to participle, and load the semantic knowledge of word after participle.When pending statement is carried out to participle, according to field dictionary and universaling dictionary, pending statement is carried out to participle.
S2, according to word segmentation result, identify the semantic node of this statement.Mainly comprise following process: for the word after participle, if there is the function word senses of a dictionary entry in the semantic knowledge of word, this word is labeled as to L, if there is the verb senses of a dictionary entry in the semantic knowledge of word, is labeled as V; To all words that are labeled as L or V, carry out LV and get rid of processing; All L marks are carried out to mark according to its concept classification, and judge whether it has rear mark, if there is rear mark, the word of rear mark is also carried out to mark, according to above-mentioned all mark generative semantics nodes.
The concrete mode of said process is as follows:
Each word is carried out to LV identification, if there is the function word senses of a dictionary entry in the semantic knowledge of word, this word is labeled as L, if there is the verb senses of a dictionary entry in the semantic knowledge of word, this word is labeled as V.Described semantic knowledge comprises generalized concept class and the subclass (being concept classification) thereof of word, and the concept generalized concept class of described word comprises dynamically, static state, thing, people, attribute and logic.
To all words that are labeled as L or V, carry out LV get rid of process, if before this word, have " ", " one " such word, cancel its L and V mark; If have after this word " " such word, cancel its L and V mark;
To all L marks, if the concept classification of this node is l1, its mark is revised as L1; Judge whether it has rear mark, " when ... time " in, mark after " time " be " when ", to the word of rear mark, generates a mark that is labeled as L1H; If the concept classification of this node is l0, its mark is revised as L0.
All L marks (comprising L0, L1 and L1H) and V mark, bring positional information, generate a semantic node, charge to a queue, be referred to as First ray.If generate and exceed 1 semantic node on a word, all charge to First ray.
S3, utilize semantic knowledge and word position to obtain the level of semantic node.First, the acquiescence level of all L marks and v mark is all designated as to 0, in the time occurring that two above-mentioned marks are adjacent, the level of second mark reduces one deck.Specific as follows:
To all semantic nodes in First ray, carry out the identification of LV level, the acquiescence level of all L marks and V mark is all designated as 0;
In the time that two L are adjacent, while there is L1L2, the level of L2 subtracts 1;
In the time that L is adjacent with V, while there is L1V2, the level of V2 subtracts 1;
In the time that L is adjacent with V, while there is V1L2, the level of L2 subtracts 1;
To fullstop punctuation mark, generate a semantic node, it is labeled as SST, charges to First ray.
S4, identify the semantic limit of different levels in this statement.Comprise: first, to all semantic nodes that are labeled as V, carry out the identification of core verb, generate language piece; Then,, to all semantic nodes that are labeled as L, generate language piece; Thereby, according to language piece generative semantics limit.
Concrete mode is as follows:
Generate a queue, be referred to as the second sequence;
To all semantic nodes that are labeled as V in First ray, carry out EG identification, generate language piece, it is labeled as CHK_EG, and language piece is added to the second sequence;
To all semantic nodes that are labeled as L in First ray, carry out following processing:
To all semantic nodes that are labeled as L1, generate a language piece, its mark is CHK_ABK, its reference position is the reference position of L1 node; Judge after this node whether have L1H, if had, language block end position is the end position of L1H; If there is no thereafter L1H, language block end position is that next-door neighbour's the next one is labeled as the reference position pos-1 of the semantic node of L, and language piece level is the level of semantic node, and language piece is added to the second sequence;
To all semantic nodes that are labeled as L0, generate a language piece, its mark is CHK_L0, and its reference position is the reference position of L0, and its end position is the end position of L0, and language piece level is the level of semantic node, and language piece is added to the second sequence;
To all semantic nodes that are labeled as L0, generate a language piece, its mark is CHK_GBK, its reference position is the end position pos+1 of L0, its end position is the reference position pos-1 of next-door neighbour's next language piece (its mark is CHK_EG or CHK_ABK or CHK_L0), language piece level is the level of semantic node, and language piece is added to the second sequence;
To being labeled as the semantic node of SST in First ray, generate a language piece, its mark is CHK_SST, joins the second sequence.The language piece CHK_SST, CHK_ABK, CHK_EG, the CHK_L0 that in this process, obtain are semantic limit.
In said process, EG identification refers to the identification of core verb, mainly to judge the weights size of each dynamic concept as EG by designing a series of orderly weights, this process comprises: first, eliminating can not form the word of core verb, the word that likely forms EG in statement is tentatively got rid of, comprised the ambiguous category of dynamic concept and static concept, logical concept, attribute ambiguous category and Different Dynamic concept.Then, remaining word is given different weights according to the feature of arranging in pairs or groups and word itself has, and selects core verb according to the ranking results of weights and positional information.Namely remaining candidate's word after getting rid of is all generated to EG, and give different weights according to the feature that they form or word itself has, consider weights ranking results and positional information and select the EG of a suitable word as statement.
S5, generate level semantic tree according to the semantic limit of each level.First, select root node; Then,, language piece high level, according to the order in this level, be suspended on root node, as child node; Finally, travel through all child nodes, all language pieces within the scope of each child node are as the child node of this child node, until do not have new leaf node to produce.
Layer semantics tree constructing method described in the present embodiment, mainly comprises the following steps: statement is carried out participle and loads semantic knowledge-base; According to LV rule and language rule, all nodes and the level thereof of identification statement; End of the sentence punctuation mark is generated to special node, as the root node of semantic tree; According to the nodal information of above-mentioned generation, it is merged, the semantic limit language piece of identification statement, is hung on root node using 0 grade of semantic limit language piece as child node; Travel through each child node until without the semantic limit of low level language piece, be hung on child node as leaf node.Output said structure tree can obtain Layer semantics tree.The scheme that Layer semantics tree in the present embodiment builds, all utilizes rule and method to realize to the analysis of statement.The reason that algorithm is under suspicion is, if rule description is too simple, and the result that rule produces or conflicting, or be not enough to parsing sentence.If think, complete dependent Rule provides analysis result exactly, just needs each rule can describe complicated language phenomenon, and this makes regular generality poor, writes and needs in a large number manually, does not have feasibility.For solving this contradiction, in this programme, by the identification of node and the identification of level, semantic limit and level thereof, carry out the scheduling of control law in different levels, different phase.Under this principle instructs, first need rule to carry out hierarchical classification, each rule-like only calls in the fixing level of analysis, and each rule is only paid close attention to the analysis to language phenomenon in contiguous language string, do not need to take into account the judgement to overall situation, but solve regular compatibility issue by scheduling.The strategy solving in the present embodiment has two: first avoid regular greediness coupling, make rule invocation have level, and call respective rule according to active information on each level; Secondly the result that, scheduling meeting generates rule according to the statement feature in different disposal stage is selected synthetic.Like this, both reduce the rule that needs coupling, also reduced the impact on final analysis of contradiction that Different Rule produces, strengthened the control to rule invocation with this, also made the possibility that is configured to of rule-based Layer semantics tree.
Above-mentioned structure semantic tree is under the guidance of hierarchical network of concepts theory, in the situation that there is no syntax resource, the semantic structure tree that only uses semantic information and language rule and obtain, make computing machine can enter the semantic deep layer of natural language, on the basis of understanding, complete the various processing to natural language, realized the first step of semantics comprehension on natural language.Build semantic tree, can be widely used in natural language processing field, as convenient in information retrieval, automatic abstract, mechanical translation, text classification and information filtering etc.The construction method of the semantic tree in the present embodiment, has been applied in patent documentation Chinese-English machine translation, has significantly improved readability and the accuracy of patent documentation translation.
embodiment 2:
In the present embodiment, provide a concrete Layer semantics tree constructing method, the basic procedure of this scheme is also as shown in Figure 1, the middle-level semantic tree construction method 100 of the present embodiment starts from step S110 and inputs pending statement, then in step S120, pending statement is carried out to pre-service, according to field dictionary and universaling dictionary, pending statement is carried out to participle, and load the semantic knowledge of word, semantic knowledge mainly comprises that the generalized concept class of word is V (dynamically), G (static state), W (thing), P (people), U (attribute), some subclasses under the large generalized concept class of L (logic) six and its pool, secondly, in step S130, identify the semantic node of this statement and its level is distinguished, the first step is to the result after participle, adopts all semantic nodes of LV rule identification, second step is to utilize semantic knowledge and word position, relatively judges the level of node, again, in step S140, identify the semantic limit of the different levels of this statement, the recognition result of the semantic node of minor sentence aspect, be identified as the semantic limit of minor sentence aspect, the recognition result of the semantic node of language piece aspect, be identified as the semantic limit of language piece aspect, then, in step S150, generate level semantic tree, according to the recognition result on semantic limit, be created on by different level on tree construction according to scheduling, finally, in step S160, export the Layer semantics tree of pending statement.
Fig. 2 is the schematic diagram of explanation node recognition 300.As shown in Figure 2, the entrance S310 of node recognition is the word segmentation result of pending language material.In step S311, word and punctuate are treated with a certain discrimination.For word, need to load to each word the semantic knowledges such as concept classification.Semantic knowledge simply comprises following two aspects: Words ' Attributes, and whether it comprises generalized concept class GCC, concept classification CC, LV attribute LV, morpheme QH, is pure V verb CHUNV; Sentence generic attribute, it comprises broad sense effect sentence GXGY, subject number of blocks GBK_NUM, piece expands a sentence EPER, GBK2 prototype sentence is sloughed off GBK2_YT, passive voice ALL_PASS, whether formed bidirectional relationship sentence R0, whether constituent ratio judges a JD0.It should be noted that the classification of concept classification and being described as follows shown in table:
Wherein the basic format of knowledge base style is as follows:
Morphology
$Feature[Value]$
For example:
Semiconductor element
$GCC[W]CC[pw]$
Represent
$CC[v]SC_GXY[GX]EPER[Y]GBK_NUM[3；4]SC_GBK1_PP[Y]$
Wherein, GCC[W] represent that the large class of concept of this entry (" semiconductor element ") is thing W, CC[pw] represent that concept classification is artificiality PW; CC[v] represent that the concept classification of this entry (" expression ") is verb, SC_GXY[GX] represent it is broad sense effect sentence, EPER[Y] represent it is that piece expands sentence, GBK_NUM[3; 4] represent it is three main or four main sentences, SC_GBK1_PP[Y] represent that GBK1 must be people or life entity.
For punctuate, fullstop will generate special semantic node, is labeled as SST, as root node.
In step S330, each word is carried out to " LV " identification, if there is logic l concept in the semantic knowledge of word, generative semantics node, this word is labeled as L, if there is dynamic v concept etc. in the semantic knowledge of word, generative semantics node, this word is labeled as V.Meanwhile, respectively the word that is labeled as V and L is carried out to ambiguous category eliminating processing by corresponding some row's discrimination rules.To all words that are labeled as V, can by as below two rules be example carry out ambiguous category get rid of process: for the word that is labeled as V, if before this word, have " ", " one " such word, cancel its V mark; If have after this word " " such word, cancel its V mark.
In step S340, to all L marks, if the concept classification of this word is l1, its mark is revised as L1; Judge whether it has rear mark, if there is rear mark, to the word of rear mark, generate a mark that is labeled as L1H.As in Chinese " when ... time ", wherein, " when " concept classification be l1, its mark can be revised as L1, and " time " be " when " after mark, " time " be labeled as L1H.If the concept classification of this word is l0, its mark is revised as L0, as Chinese " " word.
Step S350 is all nodes that identify.
In step S360, all nodes are carried out the identification of LV level and are distinguished the LEVEL information of node.To all semantic nodes in First ray, carry out the identification of LV level, it comprises following operation: the acquiescence level of all L marks and V mark is all designated as 0; In the time that two L are adjacent, while there is L1, L2, the level of L2 subtracts 1, as " that this mathematics book is on the shelf lifted down ", wherein " " and " " be two adjacent L concepts, now, " " be L1, its level is 0; And " " be L2, its level is-1; In the time that L is adjacent with V, while there is L1, V2, the level of V2 subtracts 1, as " that this mathematics book being positioned on bookshelf is lifted down ", wherein " " and " being positioned at " be two adjacent L and V concept, now, " " be L1, its level is 0; And " being positioned at " is V2, its level is-1; In the time that V is adjacent with L, while there is V1, L2, the level of L2 subtracts 1, as " applying the module relevant with user ", wherein " application " and "AND" are two adjacent V and L concept, and now, " application " is V1, its level is 0, and "AND" is L2, and its level is-1.
In step S370, the result obtaining be exactly this statement differentiation all nodes of LEVEL information, and charge to First ray, be referred to as First ray: all L marks (comprising L0, L1 and L1H) and V mark, bring the positional information in statement, as semantic node, charge to First ray; If generate and exceed 1 semantic node on a word, all charge to First ray; To punctuation mark, the semantic node SST of generation, also together charges to First ray.
Fig. 3 is the schematic diagram of declarative semantics limit identification 400.As shown in Figure 4, the entrance of semantic limit identification is all nodes and level LEVEL information thereof.
First first generate a queue, be referred to as the second sequence.
In step S410, to all semantic nodes that are labeled as V in First ray, carry out EG identification, generate language piece, it is labeled as CHK_EG, and language piece is added to the second sequence.
As " the various device that the present invention can fast access be docked with electronic equipment 10." in " access, docking " be the semantic node that is labeled as V; by language rule, two semantic nodes be weighted and fall power; " access " is by " can, fast " two word weightings; and " docking " by be close to thereafter " " it is fallen to power; in this, " access " weights are higher; be chosen as the EG of minor sentence, be labeled as CHK_EG.
In step S420, to all semantic nodes that are labeled as L in First ray, carry out following processing:
To all semantic nodes that are labeled as L1, generate a language piece, its mark is CHK_ABK, its reference position is the reference position of L1 node; Judge after this node whether have L1H, if had, language block end position is the end position of L1H; If there is no thereafter L1H, language block end position is that next-door neighbour's the next one is labeled as the reference position pos-1 of the semantic node of L, and language piece level is the level of semantic node, and language piece is added to the second sequence.
The generation situation of the CHK_ABK of following example explanation minor sentence aspect: as " storer 130 can be separated by different way."; wherein " with " be the semantic node that is labeled as L1; be not labeled as thereafter the semantic node of L1, L1H; can generate a language piece that is labeled as CHK_ABK; its reference position be " with " reference position of semantic node; end position is that the reference position of CHK_EG does not comprise this position, and this CHK_ABK language piece is " by different way "; As " the present invention removes weeds with blade in spiral rolling mode."; wherein " use " is the semantic node that is labeled as L1; have be thereafter labeled as L1 semantic node " with "; can generate a language piece that is labeled as CHK_ABK; its reference position is the reference position of " use " semantic node, end position be " with " reference position do not comprise this position, first CHK_ABK language piece of this sentence is " with blade "; the same, " in spiral rolling mode " is also this CHK_ABK; And for example " on electronic equipment 10, present media content ", wherein, " " be the semantic node that is labeled as L1, have be thereafter labeled as L1H semantic node " on ", can generate a language piece that is labeled as CHK_ABK, its reference position be " with " reference position of semantic node, end position is the position of " interior " semantic node, this CHK_ABK language piece is " on electronic equipment 10 ".The L1 of above-mentioned three examples and L1H are minor sentence aspects, and the level that its level is defaulted as 0, CHK_ABK is also 0.The generation situation of the CHK_ABK of following example explanation language piece inside, at the sentence " media content that user's Internet access presents by operating system 137." in; " access " is the CHK_EG of sentence; " media content presenting by operating system 137 " is a CHK_GBK language piece; it " presents media content by operating system 137 " by sentence, and degradation degrades; wherein " presenting " is the V semantic node of this CHK_GBK language piece; can generate CHK_EG, and its level is-1; Wherein " passing through " is the semantic node that is labeled as L1, and its level is-1, and " by operating system 137 " can generate a language piece that is labeled as CHK_ABK.Equally, be-1 at the level of the inner CHK_ABK generating of GBK language piece.
To all semantic nodes that are labeled as L0, generate a language piece, its mark is CHK_L0, and its reference position is the reference position of L0, and its end position is the end position of L0, and language piece level is the level of semantic node, and language piece is added to the second sequence.The generation situation of the CHK_L0 of following example explanation minor sentence aspect, at sentence, " user name and/or password combination are input to user interface 150 and/or authenticating device 70 by user." in, " by " being marked as L0, its hierarchical information is 0, and being generated a mark is CHK_L0 language piece, and reference position and end position are all L0; The generation situation of the CHK_L0 of following example explanation language piece aspect, in language piece " media content of being accessed by user ", " by " being marked as L0, its hierarchical information is-1, being generated a mark is CHK_L0 language piece, and reference position and end position are all L0.
To being labeled as the semantic node of SST in First ray, generate a language piece, its mark is CHK_SST, joins the second sequence.
In step S430, utilize the relation between all language piece CHK_L0 and CHK_ABK and CHK_EG, generate a language piece, its mark is CHK_GBK, its reference position is the end position pos+1 of CHK_L0, its end position is the reference position pos-1 of next-door neighbour's next language piece (its mark is CHK_ABK or CHK_EG), and language piece level is the level of semantic node, and language piece is added to the second sequence.As in above-mentioned example, " user name and/or password combination are input to user interface 150 and/or authenticating device 70 by user." in, " general " generate language piece CHK_L0, " being input to " generates language piece CHK_EG, and " user ", " user name and/or password combination " and " user interface 150 and/or authenticating device 70 " they are CHK_GBK language pieces.
In step S440, CHK_EG, CHK_ABK, CHK_L0, the CHK_SST obtaining be all semantic limits.
Determine taking SST as root node, the first level CHK_EG, CHK_L0, CHK_ABK, CHK_GBK are its child node and are hung under it, the child node of CHK_EG, the CHK_L0 of the second level, CHK_ABK, CHK_GBK stone node is hung under it, by that analogy, until be all leaf node.
embodiment 3:
In the present embodiment, provide a concrete application example, Fig. 4 and Fig. 5 are the schematic diagram that the Layer semantics tree of illustrated example statement builds result.As shown in Figure 4, pending statement is that " web browser uses URL(uniform resource locator) that HTML request is sent to the server by system control."; the semantic tree structure of minor sentence aspect is: GBK1 " web browser "+ABK " use URL(uniform resource locator) "+L0 " general "+GBK2 " HTML request " " send to "+GBK3 of+EG " by the server of system control "; wherein, CHK_SST (fullstop) language piece is as root node.The semantic node of the first level is L1 (use), L0 (general), V (sending to), and three levels are all 0; The semantic limit of the first level be CHK_ABK (use URL(uniform resource locator)), CHK_L0 (general), CHK_EG (sending to), CHK_GBK (web browser, HTML request, remote server), six language piece levels are all 0, and its child node as root node is hung up." send to " action that represents transmission according to CHK_EG, can determine that the semantic role of CHK_GBK is as follows: " web browser " is actor GBK1, " HTML request " is content GBK2, and " remote server " is target ground GBK3.GBK1, GBK2 language piece aspect semantic relation are fairly simple, although " browser " is the Semantic center of " web browser ", " request " is the Semantic center of " HTML request ", " server " is the Semantic center of " remote server ", but because there is no the semantic limit of language piece level, the word of language piece is all hung up as leaf node.The semantic tree structure of language piece aspect: L0 in GBK3 " by "+GBK2 " system "+EG " control "+CHK_L1 " "+GBK3 " server ", wherein GBK3 language piece is as root node.The semantic node of the second level be L0 (by), V (control), L1 (), three levels are all-1; The semantic limit of the second level be CHK_L0 (by), CHK_EG (control), CHK_L1 (), CHK_GBK (system, server), five language piece levels are all-1, it is hung up as child node.According to CHK_EG, " control " is the concept that represents the effect of broad sense, can determine that the semantic role of CHK_GBK is as follows: " system " is actor GBK1, and " server " is content GBK2.The semantic tree that in the present embodiment, this statement is set up as shown in Figure 4 and Figure 5.
embodiment 4:
In the present embodiment, provide a kind of system that realizes the Layer semantics tree constructing method described in above-described embodiment, the Layer semantics tree constructing system 500 in the present embodiment, structured flowchart as shown in Figure 6, comprises
Pretreatment unit S520: input pending statement, pending statement is carried out to participle, and load the semantic knowledge of word after participle;
First ray generation unit S530: according to word segmentation result, identify the semantic node of this statement; Utilize semantic knowledge and word position to obtain the level of semantic node;
The second sequence generating unit S540: the semantic limit of identifying different levels in this statement;
Layer semantics tree generation unit S550: generate level semantic tree according to the semantic limit of each level.
In addition,, in the time implementing, also comprise read statement unit and Layer semantics tree output unit S560.
Preferably, in described pretreatment unit S520, when pending statement is carried out to participle, according to field dictionary and universaling dictionary, pending statement is carried out to participle.In the present embodiment, described semantic knowledge comprises generalized concept class and the subclass thereof of word, and the generalized concept class of described word comprises dynamically, static state, thing, people, attribute, logic.
Preferably, in First ray generation unit S530, comprising:
The first subelement: for the word after participle, if having logical concept in the semantic knowledge of word,
This word is labeled as to L, if having dynamic concept in the semantic knowledge of word, is labeled as V;
The second subelement: to all words that are labeled as L or V, carry out LV and get rid of processing;
The 3rd subelement: all L marks are carried out to mark according to its concept classification, and judge whether it has rear mark, if there is rear mark, the word of rear mark is labeled as to L1H, according to above-mentioned all mark generative semantics nodes.
Also comprise the 4th subelement: using end of the sentence punctuate generative semantics node as root node.
First ray generation unit S530 also comprises:
The 5th subelement: the acquiescence level of all L marks and v mark is all designated as 0, in the time occurring that two above-mentioned marks are adjacent, it is-1 that the level of second mark reduces one deck.
The second sequence generating unit S540 comprises:
Core verb recognition unit: to all semantic nodes that are labeled as V, carry out the identification of core verb, generate language piece;
Language piece generation unit: to all semantic nodes that are labeled as L, generate language piece;
Semantic limit generation unit: according to language piece generative semantics limit.
In core verb recognition unit, carry out the identification of core verb, also comprise:
Get rid of subelement: eliminating can not form the word of core verb;
Chooser unit: remaining word is given different weights according to the feature forming and word itself has, selects core verb according to the ranking results of weights and positional information.
Layer semantics tree generation unit S550, comprising:
Root node subelement: select root node;
Child node subelement: language piece high level, according to the order in this level, be suspended on root node, as child node;
Traversal subelement: travel through all child nodes, all language pieces within the scope of each child node are as the child node of this child node, until do not have new child node to produce.
Fig. 6 is the schematic diagram of the Layer semantics tree constructing system 500 in the explanation embodiment of the present invention.Layer semantics tree structure equipment 500 comprises five unit: pretreatment unit S520, First ray generation unit S530, the second sequence generating unit S540, Layer semantics tree generation unit S550 and Layer semantics tree output unit S560.Step S510 represents the input of statement, generally refers to a complete sentence, but not sentence group or chapter.Pretreatment unit S520 comprises the processing of statement being carried out to the special punctuates such as word segmentation processing, bracket to paired, quotation marks, punctuation marks used to enclose the title, loads semantic knowledge-base, to the numeral occurring in statement and english abbreviation bind and load its semantic information, to comma, colon, pause mark, fullstop etc. effectively punctuate process and load its semantic information and adopt disambiguation rule to carry out disambiguation processing to the word of ambiguous category, the operation fundamental purpose of pretreatment unit is to get rid of to disturb to make the more succinct easily row of follow-up identification step.First ray generation unit S530 takes LV principle to process to identify all semantic node L/V to the word of all l of containing or v concept, and the position relationship that utilizes LV semantic node to present is distinguished its level, acquiescence is all 0, and it represents the first level, and the second level is-1; According to the effectively semantic node of the semantic information identification punctuate class of punctuate such as comma, colon, pause mark, fullstop.The second sequence generating unit S540 is mainly according to all semantic node L/V/SST and level recognition node limit CHK_EG, CHK_L0, CHK_ABK, CHK_GBK and level thereof.Layer semantics tree generation unit S550 is mainly the analysis for CHK_GBK inner structure, according to wherein the degrade semantic structure of sentence and other of semantic structure, above-mentioned degradation side by side of language piece internal combination symbol identification.Especially, it should be noted that, degrade identification and the minor sentence aspect of sentence of degradation is similar, and the hierarchical information of different is CHK_ABK, CHK_L0, CHK_EG is-1.Layer semantics tree output unit S560 is mainly the Layer semantics tree of being exported to obtain according to the result of Layer semantics tree generation unit, specifically comprise: determine taking SST as root node, the first level CHK_EG, CHK_L0, CHK_ABK, CHK_GBK are its child node and are hung under it, the child node of CHK_EG, the CHK_L0 of the second level, CHK_ABK, CHK_GBK stone node is hung under it, by that analogy, until be all leaf node.
Obviously, above-described embodiment is only for example is clearly described, and the not restriction to embodiment.For those of ordinary skill in the field, can also make other changes in different forms on the basis of the above description.Here without also giving exhaustive to all embodiments.And the apparent variation of being extended out thus or variation are still among the protection domain in the invention.
Claims (10)
1. for Layer semantics tree constructing method and the system of language understanding, it is characterized in that, comprise the steps:
S1, input pending statement, pending statement is carried out to participle, and load the semantic knowledge of word after participle;
S2, according to word segmentation result, identify the semantic node of this statement;
S3, utilize semantic knowledge and word position and collocation to obtain the level of semantic node;
S4, identify the semantic limit of different levels in this statement;
S5, generate level semantic tree according to the semantic limit of each level.
2. Layer semantics tree constructing method according to claim 1, is characterized in that, comprising: in described step S1, when pending statement is carried out to participle, according to field dictionary and universaling dictionary, pending statement is carried out to participle.
3. Layer semantics tree constructing method according to claim 1 and 2, is characterized in that, described semantic knowledge comprises generalized concept class and the subclass thereof of word, and the generalized concept class of described word comprises dynamically, static state, thing, people, attribute, logic.
4. according to the arbitrary described Layer semantics tree constructing method of claim 1-3, it is characterized in that, the process that " according to word segmentation result, identifies the semantic node of this statement " in described step S2, comprising:
For the word after participle, if having logical concept in the semantic knowledge of word, this word is labeled as to L, if having dynamic concept in the semantic knowledge of word, be labeled as V;
To all words that are labeled as L or V, carry out LV and get rid of processing;
All L marks are carried out to mark according to its concept classification, and judge whether it has rear mark, if there is rear mark, the word of rear mark is labeled as to L1H, according to above-mentioned all mark generative semantics nodes.
5. according to the arbitrary described Layer semantics tree constructing method of claim 1-4, it is characterized in that, the process that " according to word segmentation result, identifies the semantic node of this statement " in described step S2, also comprises: using end of the sentence punctuate generative semantics node as root node.
6. according to the arbitrary described Layer semantics tree constructing method of claim 1-5, it is characterized in that, the process of " utilizing semantic knowledge and word position and collocation to obtain the level of semantic node " in described step S3, comprising:
The acquiescence level of all L marks and v mark is all designated as 0, and in the time occurring that two above-mentioned marks are adjacent, it is-1 that the level of second mark reduces one deck.
7. according to the arbitrary described Layer semantics tree constructing method of claim 1-6, it is characterized in that, the process of " identifying the semantic limit of different levels in this statement " in described step S4, comprises
To all semantic nodes that are labeled as V, carry out the identification of core verb, generate language piece;
To all semantic nodes that are labeled as L, generate language piece;
According to language piece generative semantics limit.
8. according to the arbitrary described Layer semantics tree constructing method of claim 1-7, it is characterized in that, described in carry out the identification of core verb process comprise:
Eliminating can not form the word of core verb;
Remaining word is given different weights according to the feature forming and word itself has, and selects core verb according to the ranking results of weights and positional information.
9. according to the arbitrary described Layer semantics tree constructing method of claim 1-8, it is characterized in that, the described process that generates level semantic tree according to the semantic limit of each level, comprising:
Select root node;
Language piece high level, according to the order in this level, be suspended on root node, as child node;
Travel through all child nodes, all language pieces within the scope of each child node are as the child node of this child node, until do not have new child node to produce.
10. the Layer semantics tree constructing system that the Layer semantics tree constructing method described in claim 1-9 is corresponding, is characterized in that, comprising:
Pretreatment unit: input pending statement, pending statement is carried out to participle, and load the semantic knowledge of word after participle;
First ray generation unit: according to word segmentation result, identify the semantic node of this statement; Utilize semantic knowledge and word position and collocation to obtain the level of semantic node;
The second sequence generating unit: the semantic limit of identifying different levels in this statement;
Layer semantics tree generation unit: generate level semantic tree according to the semantic limit of each level.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410216929.8A CN104142917B (en) | 2014-05-21 | 2014-05-21 | A kind of Layer semantics tree constructing method and system for language understanding |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410216929.8A CN104142917B (en) | 2014-05-21 | 2014-05-21 | A kind of Layer semantics tree constructing method and system for language understanding |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104142917A true CN104142917A (en) | 2014-11-12 |
| CN104142917B CN104142917B (en) | 2018-05-01 |
Family
ID=51852093
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410216929.8A Expired - Fee Related CN104142917B (en) | 2014-05-21 | 2014-05-21 | A kind of Layer semantics tree constructing method and system for language understanding |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104142917B (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104657348A (en) * | 2015-02-06 | 2015-05-27 | 北京师范大学 | Method and system for extracting predicate chunk in Chinese sentence |
| CN105068995A (en) * | 2015-08-19 | 2015-11-18 | 刘战雄 | Natural language semantic calculation method and apparatus based on question semantics |
| CN106021286A (en) * | 2016-04-29 | 2016-10-12 | 东北电力大学 | Method for language understanding based on language structure |
| CN106897371A (en) * | 2017-01-18 | 2017-06-27 | 南京云思创智信息科技有限公司 | Chinese text classification system and method |
| CN107679035A (en) * | 2017-10-11 | 2018-02-09 | 石河子大学 | A kind of information intent detection method, device, equipment and storage medium |
| CN108874791A (en) * | 2018-07-06 | 2018-11-23 | 北京联合大学 | A kind of semantic analysis based on minimum semantic chunk and Chinese-English sequence adjusting method and system |
| CN109271621A (en) * | 2017-07-18 | 2019-01-25 | 腾讯科技（北京）有限公司 | Semanteme disambiguates processing method, device and its equipment |
| CN109446719A (en) * | 2018-11-15 | 2019-03-08 | 北京神舟航天软件技术有限公司 | A kind of implementation method of the customized product tree of element type |
| CN109815490A (en) * | 2019-01-04 | 2019-05-28 | 平安科技（深圳）有限公司 | Text analyzing method, apparatus, equipment and storage medium |
| CN110192203A (en) * | 2016-11-03 | 2019-08-30 | 易享信息技术有限公司 | Joint multitask neural network model for multiple natural language processings (NLP) task |
| CN110209829A (en) * | 2018-02-12 | 2019-09-06 | 百度在线网络技术（北京）有限公司 | Information processing method and device |
| CN110276080A (en) * | 2019-06-28 | 2019-09-24 | 第四范式（北京）技术有限公司 | A kind of semantic processes method and system |
| CN111814487A (en) * | 2020-07-17 | 2020-10-23 | 科大讯飞股份有限公司 | Semantic understanding method, device, equipment and storage medium |
| WO2021135103A1 (en) * | 2020-05-29 | 2021-07-08 | 平安科技（深圳）有限公司 | Method and apparatus for semantic analysis, computer device, and storage medium |
| CN113128226A (en) * | 2019-12-31 | 2021-07-16 | 阿里巴巴集团控股有限公司 | Named entity recognition method and device, electronic equipment and computer storage medium |
| CN115017913A (en) * | 2022-04-21 | 2022-09-06 | 广州世纪华轲科技有限公司 | Semantic component analysis method based on master-slave framework mode |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5424947A (en) * | 1990-06-15 | 1995-06-13 | International Business Machines Corporation | Natural language analyzing apparatus and method, and construction of a knowledge base for natural language analysis |
| CN1430160A (en) * | 2001-12-30 | 2003-07-16 | 中国科学院声学研究所 | Expression and obtaining method of natural language sentence cluster |
| CN101059806A (en) * | 2007-06-06 | 2007-10-24 | 华东师范大学 | Word sense based local file searching method |
| US20090077113A1 (en) * | 2005-05-12 | 2009-03-19 | Kabire Fidaali | Device and method for semantic analysis of documents by construction of n-ary semantic trees |
-
2014
- 2014-05-21 CN CN201410216929.8A patent/CN104142917B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5424947A (en) * | 1990-06-15 | 1995-06-13 | International Business Machines Corporation | Natural language analyzing apparatus and method, and construction of a knowledge base for natural language analysis |
| CN1430160A (en) * | 2001-12-30 | 2003-07-16 | 中国科学院声学研究所 | Expression and obtaining method of natural language sentence cluster |
| US20090077113A1 (en) * | 2005-05-12 | 2009-03-19 | Kabire Fidaali | Device and method for semantic analysis of documents by construction of n-ary semantic trees |
| CN101059806A (en) * | 2007-06-06 | 2007-10-24 | 华东师范大学 | Word sense based local file searching method |
Non-Patent Citations (2)
| Title |
|---|
| ZHU XIAOJIAN，JIN YAOHONG: "Hierarchical Semantic-Category-Tree Model for Chinese－English Machine Translation", 《中国通信》 * |
| 刘智颖，郭艳波，晋耀红: "汉英机器翻译中格式转换研究", 《计算机工程与应用》 * |
Cited By (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104657348B (en) * | 2015-02-06 | 2017-11-14 | 北京师范大学 | Predicate language block extracting method and system in a kind of Chinese sentence |
| CN104657348A (en) * | 2015-02-06 | 2015-05-27 | 北京师范大学 | Method and system for extracting predicate chunk in Chinese sentence |
| CN105068995B (en) * | 2015-08-19 | 2018-05-29 | 刘战雄 | A kind of method and device of the natural language semantic computation based on query semanteme |
| CN105068995A (en) * | 2015-08-19 | 2015-11-18 | 刘战雄 | Natural language semantic calculation method and apparatus based on question semantics |
| CN107562731B (en) * | 2015-08-19 | 2020-09-04 | 刘战雄 | Natural language semantic calculation method and device based on question semantics |
| CN107562731A (en) * | 2015-08-19 | 2018-01-09 | 刘战雄 | A kind of method and device of the natural language semantic computation semantic based on query |
| CN106021286A (en) * | 2016-04-29 | 2016-10-12 | 东北电力大学 | Method for language understanding based on language structure |
| US11797825B2 (en) | 2016-11-03 | 2023-10-24 | Salesforce, Inc. | Training a joint many-task neural network model using successive regularization |
| US11783164B2 (en) | 2016-11-03 | 2023-10-10 | Salesforce.Com, Inc. | Joint many-task neural network model for multiple natural language processing (NLP) tasks |
| CN110192203A (en) * | 2016-11-03 | 2019-08-30 | 易享信息技术有限公司 | Joint multitask neural network model for multiple natural language processings (NLP) task |
| CN110192204A (en) * | 2016-11-03 | 2019-08-30 | 易享信息技术有限公司 | The deep neural network model of data is handled by multiple language task levels |
| CN110192204B (en) * | 2016-11-03 | 2023-09-29 | 硕动力公司 | Deep neural network model for processing data through multiple language task hierarchies |
| CN110192203B (en) * | 2016-11-03 | 2023-10-03 | 硕动力公司 | Joint multitasking neural network model for multiple Natural Language Processing (NLP) tasks |
| CN106897371A (en) * | 2017-01-18 | 2017-06-27 | 南京云思创智信息科技有限公司 | Chinese text classification system and method |
| CN106897371B (en) * | 2017-01-18 | 2020-04-21 | 南京云思创智信息科技有限公司 | Chinese text classification system and method |
| CN109271621A (en) * | 2017-07-18 | 2019-01-25 | 腾讯科技（北京）有限公司 | Semanteme disambiguates processing method, device and its equipment |
| CN109271621B (en) * | 2017-07-18 | 2023-04-18 | 腾讯科技（北京）有限公司 | Semantic disambiguation processing method, device and equipment |
| CN107679035B (en) * | 2017-10-11 | 2020-06-12 | 石河子大学 | Information intention detection method, device, equipment and storage medium |
| CN107679035A (en) * | 2017-10-11 | 2018-02-09 | 石河子大学 | A kind of information intent detection method, device, equipment and storage medium |
| CN110209829A (en) * | 2018-02-12 | 2019-09-06 | 百度在线网络技术（北京）有限公司 | Information processing method and device |
| CN110209829B (en) * | 2018-02-12 | 2021-06-29 | 百度在线网络技术（北京）有限公司 | Information processing method and device |
| CN108874791A (en) * | 2018-07-06 | 2018-11-23 | 北京联合大学 | A kind of semantic analysis based on minimum semantic chunk and Chinese-English sequence adjusting method and system |
| CN108874791B (en) * | 2018-07-06 | 2022-05-24 | 北京联合大学 | Semantic analysis and Chinese-English sequencing method and system based on minimum semantic block |
| CN109446719A (en) * | 2018-11-15 | 2019-03-08 | 北京神舟航天软件技术有限公司 | A kind of implementation method of the customized product tree of element type |
| CN109815490A (en) * | 2019-01-04 | 2019-05-28 | 平安科技（深圳）有限公司 | Text analyzing method, apparatus, equipment and storage medium |
| CN109815490B (en) * | 2019-01-04 | 2023-11-14 | 平安科技（深圳）有限公司 | Text analysis method, device, equipment and storage medium |
| CN110276080A (en) * | 2019-06-28 | 2019-09-24 | 第四范式（北京）技术有限公司 | A kind of semantic processes method and system |
| CN110276080B (en) * | 2019-06-28 | 2023-10-17 | 第四范式（北京）技术有限公司 | Semantic processing method and system |
| CN113128226A (en) * | 2019-12-31 | 2021-07-16 | 阿里巴巴集团控股有限公司 | Named entity recognition method and device, electronic equipment and computer storage medium |
| WO2021135103A1 (en) * | 2020-05-29 | 2021-07-08 | 平安科技（深圳）有限公司 | Method and apparatus for semantic analysis, computer device, and storage medium |
| CN111814487A (en) * | 2020-07-17 | 2020-10-23 | 科大讯飞股份有限公司 | Semantic understanding method, device, equipment and storage medium |
| CN115017913A (en) * | 2022-04-21 | 2022-09-06 | 广州世纪华轲科技有限公司 | Semantic component analysis method based on master-slave framework mode |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104142917B (en) | 2018-05-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104142917A (en) | Hierarchical semantic tree construction method and system for language understanding | |
| Zhang et al. | Generating fluent adversarial examples for natural languages | |
| Qi et al. | Openhownet: An open sememe-based lexical knowledge base | |
| Amplayo et al. | An adaptable fine-grained sentiment analysis for summarization of multiple short online reviews | |
| Cimiano et al. | Automatic acquisition of taxonomies from text: FCA meets NLP | |
| JP2021168124A (en) | Entity linking method, device, electronic device, storage medium, and computer program | |
| Lou et al. | Emoji-based sentiment analysis using attention networks | |
| WO2017196784A1 (en) | Ontology discovery based on distributional similarity and lexico-semantic relations | |
| Mehmood et al. | A precisely xtreme-multi channel hybrid approach for roman urdu sentiment analysis | |
| Lu et al. | Sentiment analysis of film review texts based on sentiment dictionary and SVM | |
| US11138506B2 (en) | Abstraction and portability to intent recognition | |
| Marreddy et al. | Clickbait detection in telugu: Overcoming nlp challenges in resource-poor languages using benchmarked techniques | |
| CN104978332A (en) | UGC label data generating method, UGC label data generating device, relevant method and relevant device | |
| Bellaachia et al. | Hg-rank: A hypergraph-based keyphrase extraction for short documents in dynamic genre | |
| Drumond et al. | Extracting ontology concept hierarchies from text using markov logic | |
| Bollegala et al. | ClassiNet--Predicting missing features for short-text classification | |
| Nikhila et al. | Text imbalance handling and classification for cross-platform cyber-crime detection using deep learning | |
| Mujahid et al. | Arabic ChatGPT Tweets Classification using RoBERTa and BERT Ensemble Model | |
| Liang et al. | Micro‐blog sentiment classification using Doc2vec+ SVM model with data purification | |
| Navigli et al. | BabelNetXplorer: a platform for multilingual lexical knowledge base access and exploration | |
| Al-Arfaj et al. | Towards ontology construction from Arabic texts-a proposed framework | |
| CN107784112A (en) | Short text data Enhancement Method, system and detection authentication service platform | |
| Mir et al. | Multi-Level Knowledge Engineering Approach for Mapping Implicit Aspects to Explicit Aspects. | |
| Munir et al. | A comparison of topic modelling approaches for urdu text | |
| Marshall et al. | The Diagrammatic AI Language (DIAL): Version 0.1 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180501 Termination date: 20210521 |