CN108256077B - Dynamic extended knowledge graph reasoning method oriented to China mobile intelligent customer service - Google Patents

Dynamic extended knowledge graph reasoning method oriented to China mobile intelligent customer service Download PDF

Info

Publication number
CN108256077B
CN108256077B CN201810049053.0A CN201810049053A CN108256077B CN 108256077 B CN108256077 B CN 108256077B CN 201810049053 A CN201810049053 A CN 201810049053A CN 108256077 B CN108256077 B CN 108256077B
Authority
CN
China
Prior art keywords
knowledge
knowledge graph
atoms
constructing
semantic
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201810049053.0A
Other languages
Chinese (zh)
Other versions
CN108256077A (en
Inventor
李鹏华
刘太林
朱智勤
李嫄源
朱庆元
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chongqing University of Post and Telecommunications
Original Assignee
Chongqing University of Post and Telecommunications
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Chongqing University of Post and Telecommunications filed Critical Chongqing University of Post and Telecommunications
Priority to CN201810049053.0A priority Critical patent/CN108256077B/en
Publication of CN108256077A publication Critical patent/CN108256077A/en
Application granted granted Critical
Publication of CN108256077B publication Critical patent/CN108256077B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
    • G06F16/367—Ontology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/332—Query formulation
    • G06F16/3329—Natural language query formulation or dialogue systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Computational Linguistics (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Animal Behavior & Ethology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Human Computer Interaction (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention relates to a dynamic extended knowledge graph reasoning method for China mobile intelligent customer service, and belongs to the technical field of artificial intelligence. The method comprises the following steps: constructing a ternary knowledge graph under characterization learning; a knowledge graph Top-k query technology under a limit technology; single service type multi-turn dialogue scene knowledge reasoning; and (4) performing multiple rounds of dialogue scene new semantic mining across business types. The invention provides the natural language information for effective response interaction for the user, and meets the requirement of intelligent customer service for multiple rounds of conversations.

Description

Dynamic extended knowledge graph reasoning method oriented to China mobile intelligent customer service
Technical Field
The invention belongs to the technical field of artificial intelligence, and relates to a dynamic extended knowledge graph reasoning method for China mobile intelligent customer service.
Background
With the continuous development of computer natural language technology, intelligent multi-turn dialog systems are receiving wide attention. Meanwhile, intelligent customer service is developed on the basis of large-scale knowledge processing. The intelligent customer service conversation not only establishes a quick and effective communication means for enterprises and users, but also provides statistical analysis information required by refined management for the enterprises.
In the prior art, the intelligent customer service mostly adopts a keyword and key sentence matching mode, calls the stored question answers from the knowledge base, and feeds back the question answers to the user by adopting a question-answer type conversation mode. The mode can cause the conversation mode to be rigid and fixed, the humanized design is lacked, the user experience is reduced, and when the user does not clearly express the intention, the whole consultation period can be prolonged, and the customer service efficiency is reduced.
For the existing situation, a novel intelligent customer service interaction technology is urgently needed to be developed, and deep learning provides a new idea for intelligent customer service in the direction of natural language processing.
Disclosure of Invention
In view of this, the present invention provides a dynamic extended knowledge graph inference method for china mobile intelligent customer service, and aims to provide a fast and efficient customer service experience and a humanized user experience for a user.
In order to achieve the purpose, the invention provides the following technical scheme:
a dynamic extended knowledge graph inference method for intelligent customer service in china, as shown in fig. 1, specifically comprising:
101: constructing a ternary knowledge graph under characterization learning;
102: inquiring a knowledge graph Top-k under a limit technology;
103: single service type multi-turn dialogue scene knowledge reasoning;
104: and (4) performing multiple rounds of dialogue scene new semantic mining across business types.
Further, in step 101, the constructing of the ternary knowledge graph under the characterization learning includes: constructing a ternary knowledge graph by constructing a triplet, obtaining a core tensor and a factor matrix by tensor decomposition, and regarding a result restored by the core tensor and the factor matrix as the probability of establishing the corresponding triplet; constructing a learning target function L to ensure that triples appearing in a knowledge graph obtain a higher learning target value, and setting a mapping function to map head entities and tail entities of the triples into semantic spaces related to a target relationship to realize knowledge conversion;
said objective function
Figure BDA0001551831080000021
Wherein h represents a head semantic entity, r represents a semantic relationship, t represents a tail semantic entity, h 'represents a mapping head semantic entity, t' represents a mapping tail semantic entity, fr(h, t) represents an energy function.
Further, in step 101, the constructing of the ternary knowledge graph under the characterization learning further includes: and constructing a three-dimensional tensor definition knowledge graph triple attached with semantic information according to the human knowledge representation structure, generating feature vectors related to the knowledge entity and the required answer through a multi-column convolution neural network, and realizing knowledge graph reasoning in the same semantic space by using a similarity scoring mechanism.
Further, in step 102, the knowledge graph Top-k query under the bounding technique includes: calculating the association scores of the knowledge atoms u and v in the knowledge graph in real time without adopting indexes
Figure BDA0001551831080000022
Searching for optimal k embeddings by using the upper and lower bounds of the embedding cost, and deducing a Top-k query result; realizing distributed upper and lower boundary refinement and distributed termination condition check through super iteration;
the association scores of the knowledge atoms u and v in the knowledge graph are as follows:
Figure BDA0001551831080000023
wherein lw,vAnd nu,vRespectively the length and the number of the shortest path between the knowledge atom u and the knowledge atom v; α is a predefined constant, which takes a value between 0 and 1; n is a value less than
Figure BDA0001551831080000024
A constant of (d); when n isu,vAt > N, the upper bound of the relevance score between u and v is
Figure BDA0001551831080000025
Further, in step 102, the knowledge graph Top-k query under the bounding technique further includes: constructing a cost function C (f), and selecting K for each atom to be checked*A candidate knowledge atom; combining the candidate knowledge atoms of each atom to be checked to obtain a candidate embedded set; by using breadth-first search algorithm, on the basis of the association score and the number of times each breadth-first search iterates, an upper bound calculation function is constructed
Figure BDA0001551831080000026
And lower bound calculation function
Figure BDA0001551831080000027
Refining the upper and lower bounds of the association value between the knowledge atom v and the source point s;
the cost function is as follows:
Figure BDA0001551831080000028
wherein C (f) is the embedding cost of embedding f,
Figure BDA0001551831080000029
in order for a collection of atoms to be known,
Figure BDA00015518310800000210
to clarify the set of knowledge atoms, qiAnd q isjAre all query knowledge atoms, f (q)i) And f (q)j) Are all matching knowledge atoms; for the first part of the formula
Figure BDA0001551831080000031
Each query knowledge atom qiConsider qiMatching knowledge atom f (q)i) Relating to the determined knowledge atom; for the second part of the formula
Figure BDA0001551831080000032
Consider f (q)i) The relation between the knowledge atoms to be checked and the matching knowledge atoms;
the upper bound calculation function:
Figure BDA0001551831080000033
the lower bound calculation function:
Figure BDA0001551831080000034
wherein the content of the first and second substances,
Figure BDA0001551831080000035
indicating that the breadth-first search visited v in the first t-1 iterations,
Figure BDA0001551831080000036
indicating that the breadth-first search has not visited v in the previous t-1 iterations.
Further, in step 103, the single-service-type multi-turn dialog scenario knowledge inference includes: in the ternary knowledge graphs belonging to different application scenes, possible combination sets and score functions of all knowledge atoms are constructed, and the degree of closeness of association among the knowledge atoms is judged according to the magnitude of function values, so that knowledge inference and new semantic mining of the knowledge graph with a single scene are realized.
Further, in step 104, the mining of new semantics across the service type multi-turn dialog scenario includes: constructing a score function to obtain a global score value; representing elements and corresponding path vectors and type vectors by adopting weight vectors of a first convolution layer in a multi-column convolution neural network, thereby obtaining three characteristic vectors of answer types, answer paths and entities around the answers; and judging the degree of closeness of the association between the cross-scene knowledge atoms according to the score, thereby realizing the knowledge inference and the new semantic mining of the cross-scene knowledge graph.
Furthermore, on the basis of single-service-type multi-round conversation scene knowledge reasoning, multi-scene knowledge scores are fused, and cross-service-type multi-round conversation scene new semantic mining is achieved.
The invention has the beneficial effects that: the invention can effectively respond to the natural language information of user interaction and meet the requirement of intelligent customer service for multiple rounds of conversations.
Drawings
In order to make the object, technical scheme and beneficial effect of the invention more clear, the invention provides the following drawings for explanation:
FIG. 1 is a schematic diagram of a dynamic extended knowledge graph reasoning process oriented to mobile intelligent customer service according to the present invention;
FIG. 2 is a schematic representation of the ternary knowledge characterization of the present invention;
FIG. 3 is a diagram of a distributed query scheme and super-iteration process of the present invention;
FIG. 4 is a depth knowledge reasoning process under the cross-boundary fusion of the present invention.
Detailed Description
Preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
The invention provides a dynamic extended knowledge graph reasoning method facing China mobile intelligent customer service, which constructs a semantic knowledge graph by constructing triples (h, r, t), as shown in figure 2, wherein h represents a head semantic entity, r represents a semantic relation, and t represents a tail semantic entity. And obtaining a core tensor and a factor matrix through tensor decomposition, wherein each two-dimensional matrix slice in the core tensor represents a semantic relation, and each row in the factor matrix represents a semantic entity. The result of the restoration from the core tensor and factor matrix can be seen as the probability that the corresponding triplet holds.
Assuming that the vectors obtained after h and t are subjected to some kind of mapping related to r are similar or equal, the energy function f is definedrOn the premise of (h, t), a learning objective function is constructed
Figure BDA0001551831080000041
The method can ensure that the triples appearing in the knowledge graph obtain higher learning target values, and meanwhile punishment is carried out on the triples which do not appear. The semantic relation is expressed by adopting a mapping vector or a mapping matrix, and the knowledge conversion is realized by setting a mapping function to map the head entity and the tail entity of the triple into a semantic space related to the target relation.
Calculating the association score of the knowledge atoms in the knowledge graph in real time under the condition of not adopting indexes
Figure BDA0001551831080000042
And the upper and lower boundaries of the embedding cost, searching the optimal k embedding by using the upper and lower boundaries, and deducing the query result of Top-k. Wherein lw,vAnd nu,vRespectively the length and the number of the shortest path between the knowledge atom u and the knowledge atom v; α is a predefined constant, which takes a value between 0 and 1; n is a value less than
Figure BDA0001551831080000043
Is constant. When n isu,vAt > N, the upper bound of the relevance score between u and v is
Figure BDA0001551831080000044
Constructing a cost function
Figure BDA0001551831080000045
So as to select K for each atom to be identified*A candidate knowledge atom. Wherein C (f) is the embedding cost of embedding f,
Figure BDA0001551831080000046
in order for a collection of atoms to be known,
Figure BDA0001551831080000047
to clarify the set of knowledge atoms, for each query knowledge atom q of the first partiConsider qiMatching knowledge atom f (q)i) Relating to the determined knowledge atom; for the second part, consider f (q)i) And the relation between the knowledge atoms to be checked and the matching knowledge atoms.
The invention combines the candidate knowledge atoms of each atom to be checked to obtain a candidate embedded set. By using breadth-first search algorithm, on the basis of the association score and the number of times each breadth-first search iterates, an upper bound calculation function is constructed
Figure BDA0001551831080000051
And lower bound calculation function
Figure BDA0001551831080000052
And refining the upper and lower bounds of the association value between the knowledge atom v and the source point s. Wherein the content of the first and second substances,
Figure BDA0001551831080000053
indicating that the breadth-first search visited v in the first t-1 iterations,
Figure BDA0001551831080000054
indicating that the breadth-first search has not visited v in the previous t-1 iterations.
As shown in FIG. 3, when a knowledge query is executed, the query system initiates a master subtask and a number of query subtasks. The main control subtask coordinates each inquiry subtask, and realizes distributed upper and lower boundary refinement and distributed termination condition check through super-iteration.
As shown in fig. 4, on the basis of knowledge inference of a single service conversation scene, multi-scene knowledge scores are fused to realize cross-boundary mining of new semantics of a multi-round conversation scene.
In a ternary knowledge graph belonging to different application scenes, a set A of all possible combinations of knowledge atoms (triples) is formed, and a scoring function s (h, l, t) ═ s acting on the set A is designed1(h,l,t)+s2And (h, l, t), judging the degree of closeness of the association between the knowledge atoms according to the level of the score function value, thereby realizing knowledge inference and new semantic mining of the single scene knowledge graph. Here, the first and second liquid crystal display panels are,
Figure BDA0001551831080000055
representing two-dimensional interaction items between knowledge atoms, wherein
Figure BDA0001551831080000056
Respectively embedded in head and tail entities R of (h, l, t)d1In (1),
Figure BDA0001551831080000057
and
Figure BDA0001551831080000058
is R dependent on the relation ld1D is a diagonal matrix that does not depend on the input triplets.
Figure BDA0001551831080000059
Representing three-dimensional interaction items between knowledge atoms, wherein RlIs a dimension of (d)2,d2) The matrix of (a) is,
Figure BDA00015518310800000510
and
Figure BDA00015518310800000511
respectively insert Rd2As a representation of the head-to-tail entity.
Similar to the composition of set a, expanding set boundaries into all business scenarios constitutes a new set B of possible combinations of knowledge atoms. Constructing a scoring function with the same form as the single scene scoring function to obtain a global scoring value Q1. And sharing the elements in the set B and the path embedding (vector) and type embedding attached to the elements, and respectively submitting the elements to a multi-column convolutional neural network to complete the supervised scene classification of the appointed type. Correspondingly, the weight vector of the first convolution layer in the multi-column convolution neural network is adopted to represent the elements in the B set and the corresponding path vector and type vector. Therefore, three feature vectors of answer type, answer path and entities around the answer are obtained. Similarity calculation is carried out on the three feature vectors pairwise, and the global score value Q is obtained through summation2. By making Q pair1And Q2Performing dot product operation to obtain a fused global score value Q3. According to Q3The degree of closeness of the association between the cross-scene knowledge atoms is judged according to the magnitude of the numerical value, so that the knowledge inference and the new semantic mining of the cross-scene knowledge graph are realized.
Through the above description of the embodiments, the functions, if implemented in the form of software functional units and sold or used as independent products, may be stored in a computer to execute the methods described in the various embodiments or some portions of the embodiments.
Finally, it is noted that the above-mentioned preferred embodiments illustrate rather than limit the invention, and that, although the invention has been described in detail with reference to the above-mentioned preferred embodiments, it will be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the scope of the invention as defined by the appended claims.

Claims (4)

1. A dynamic extended knowledge graph reasoning method facing China mobile intelligent customer service is characterized in that: the method specifically comprises the following steps:
s1: constructing a ternary knowledge graph under characterization learning;
s2: the knowledge graph Top-k query under the limit technology specifically comprises the following steps: calculating the association scores of the knowledge atoms u and v in the knowledge graph in real time without adopting indexes
Figure FDA0003189347060000011
Searching for optimal k embeddings by using the upper and lower bounds of the embedding cost, and deducing a Top-k query result; realizing distributed upper and lower boundary refinement and distributed termination condition check through super iteration;
the association scores of the knowledge atoms u and v in the knowledge graph are as follows:
Figure FDA0003189347060000012
wherein lw,vAnd nu,vRespectively the length and the number of the shortest path between the knowledge atom u and the knowledge atom v; α is a predefined constant, which takes a value between 0 and 1; n is a value less than
Figure FDA0003189347060000013
A constant of (d); when n isu,vAt > N, the upper bound of the relevance score between u and v is
Figure FDA0003189347060000014
Constructing a cost function C (f), and selecting K for each atom to be checked*A candidate knowledge atom; combining the candidate knowledge atoms of each atom to be checked to obtain a candidate embedded set; by using breadth-first search algorithm, on the basis of the association score and the number of times each breadth-first search iterates, an upper bound calculation function is constructed
Figure FDA0003189347060000015
And lower bound calculation function
Figure FDA0003189347060000016
Refining the upper and lower bounds of the association value between the knowledge atom v and the source point s;
the cost function is as follows:
Figure FDA0003189347060000017
wherein C (f) is the embedding cost of embedding f,
Figure FDA0003189347060000018
in order for a collection of atoms to be known,
Figure FDA0003189347060000019
to clarify the set of knowledge atoms, qiAnd q isjAre all query knowledge atoms, f (q)i) And f (q)j) Are all matching knowledge atoms; for the first part of the formula
Figure FDA00031893470600000110
Each query knowledge atom qiConsider qiMatching knowledge atom f (q)i) Relating to the determined knowledge atom; for the second part of the formula
Figure FDA00031893470600000111
Consider f (q)i) The relation between the knowledge atoms to be checked and the matching knowledge atoms;
the upper bound calculation function:
Figure FDA0003189347060000021
the lower bound calculation function:
Figure FDA0003189347060000022
wherein the content of the first and second substances,
Figure FDA0003189347060000023
indicating that the breadth-first search visited v in the first t-1 iterations,
Figure FDA0003189347060000024
representing that the breadth-first search has not visited v in the previous t-1 iterations;
s3: the knowledge inference of the single service type multi-turn conversation scene specifically comprises the following steps: constructing possible combination sets and score functions of all knowledge atoms in the ternary knowledge graphs belonging to different application scenes, and judging the degree of closeness of association among the knowledge atoms according to the height of function values, thereby realizing knowledge inference and new semantic mining of the knowledge graph of a single scene;
s4: the cross-service type multi-turn dialogue scene new semantic mining specifically comprises the following steps: constructing a score function to obtain a global score value; representing elements and corresponding path vectors and type vectors by adopting weight vectors of a first convolution layer in a multi-column convolution neural network, thereby obtaining three characteristic vectors of answer types, answer paths and entities around the answers; and judging the degree of closeness of the association between the cross-scene knowledge atoms according to the score, thereby realizing the knowledge inference and the new semantic mining of the cross-scene knowledge graph.
2. The dynamic extended knowledge graph reasoning method oriented to the intelligent customer service of China as claimed in claim 1, wherein: in step S1, the constructing of the ternary knowledge graph under the characterization learning includes: constructing a ternary knowledge graph by constructing a triplet, obtaining a core tensor and a factor matrix by tensor decomposition, and regarding a result restored by the core tensor and the factor matrix as the probability of establishing the corresponding triplet; constructing a learning target function L to ensure that triples appearing in a knowledge graph obtain a higher learning target value, and setting a mapping function to map head entities and tail entities of the triples into semantic spaces related to a target relationship to realize knowledge conversion;
said objective function
Figure FDA0003189347060000025
Wherein h represents a head semantic entity, r represents a semantic relationship, t represents a tail semantic entity, h 'represents a mapping head semantic entity, t' represents a mapping tail semantic entity, fr(h, t) represents an energy function.
3. The dynamic extended knowledge graph reasoning method oriented to the intelligent customer service of China as claimed in claim 2, wherein: in step S1, the constructing of the ternary knowledge graph under the characterization learning further includes: and constructing a three-dimensional tensor definition knowledge graph triple attached with semantic information according to the human knowledge representation structure, generating feature vectors related to the knowledge entity and the required answer through a multi-column convolution neural network, and realizing knowledge graph reasoning in the same semantic space by using a similarity scoring mechanism.
4. The dynamic extended knowledge graph reasoning method oriented to the intelligent customer service of China as claimed in claim 1, wherein: on the basis of single-service-type multi-round conversation scene knowledge reasoning, multi-scene knowledge scores are fused, and novel semantic mining of the cross-service-type multi-round conversation scenes is achieved.
CN201810049053.0A 2018-01-18 2018-01-18 Dynamic extended knowledge graph reasoning method oriented to China mobile intelligent customer service Active CN108256077B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810049053.0A CN108256077B (en) 2018-01-18 2018-01-18 Dynamic extended knowledge graph reasoning method oriented to China mobile intelligent customer service

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810049053.0A CN108256077B (en) 2018-01-18 2018-01-18 Dynamic extended knowledge graph reasoning method oriented to China mobile intelligent customer service

Publications (2)

Publication Number Publication Date
CN108256077A CN108256077A (en) 2018-07-06
CN108256077B true CN108256077B (en) 2022-02-08

Family

ID=62741074

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810049053.0A Active CN108256077B (en) 2018-01-18 2018-01-18 Dynamic extended knowledge graph reasoning method oriented to China mobile intelligent customer service

Country Status (1)

Country Link
CN (1) CN108256077B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110222162A (en) * 2019-05-10 2019-09-10 天津中科智能识别产业技术研究院有限公司 A kind of intelligent answer method based on natural language processing and knowledge mapping
CN110149228B (en) * 2019-05-20 2021-11-23 湖南友道信息技术有限公司 Top-k elephant flow prediction method and system based on discretization tensor filling

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106528609A (en) * 2016-09-28 2017-03-22 厦门理工学院 Vector constraint embedded transformation knowledge graph inference method
CN107015963A (en) * 2017-03-22 2017-08-04 重庆邮电大学 Natural language semantic parsing system and method based on deep neural network
CN107368468A (en) * 2017-06-06 2017-11-21 广东广业开元科技有限公司 A kind of generation method and system of O＆M knowledge mapping
CN107451230A (en) * 2017-07-24 2017-12-08 宗晖（上海）机器人有限公司 A kind of answering method and question answering system

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10509889B2 (en) * 2014-11-06 2019-12-17 ezDI, Inc. Data processing system and method for computer-assisted coding of natural language medical text

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106528609A (en) * 2016-09-28 2017-03-22 厦门理工学院 Vector constraint embedded transformation knowledge graph inference method
CN107015963A (en) * 2017-03-22 2017-08-04 重庆邮电大学 Natural language semantic parsing system and method based on deep neural network
CN107368468A (en) * 2017-06-06 2017-11-21 广东广业开元科技有限公司 A kind of generation method and system of O＆M knowledge mapping
CN107451230A (en) * 2017-07-24 2017-12-08 宗晖（上海）机器人有限公司 A kind of answering method and question answering system

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
"The More You Know: Using Knowledge Graphs for Image Classification";Kenneth Marino 等;《2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)》;20170726;第20-28页 *

Also Published As

Publication number Publication date
CN108256077A (en) 2018-07-06

Similar Documents

Publication Publication Date Title
CN109919316B (en) Method, device and equipment for acquiring network representation learning vector and storage medium
CN110175628A (en) A kind of compression algorithm based on automatic search with the neural networks pruning of knowledge distillation
CN108595533B (en) Article recommendation method based on collaborative filtering, storage medium and server
CN108154198A (en) Knowledge base entity normalizing method, system, terminal and computer readable storage medium
Alshmrany Adaptive learning style prediction in e-learning environment using levy flight distribution based CNN model
CN107392782A (en) Corporations' construction method, device and computer-processing equipment based on word2Vec
CN110222838B (en) Document sorting method and device, electronic equipment and storage medium
CN116756574B (en) Training method, using method, device and equipment of multi-mode pre-training model
CN116386899A (en) Graph learning-based medicine disease association relation prediction method and related equipment
CN108256077B (en) Dynamic extended knowledge graph reasoning method oriented to China mobile intelligent customer service
CN114781503A (en) Click rate estimation method based on depth feature fusion
CN116957128A (en) Service index prediction method, device, equipment and storage medium
CN115757897A (en) Intelligent culture resource recommendation method based on knowledge graph convolution network
CN113343100B (en) Smart city resource recommendation method and system based on knowledge graph
CN116484067A (en) Target object matching method and device and computer equipment
Perdih et al. Semantic reasoning from model-agnostic explanations
CN114268625B (en) Feature selection method, device, equipment and storage medium
CN114625886A (en) Entity query method and system based on knowledge graph small sample relation learning model
CN112927810A (en) Smart medical response method based on big data and smart medical cloud computing system
TWI419071B (en) Active knowledge management system, method and computer program product for problem solving
Zhang et al. [Retracted] Cloud Application in the Construction of English Virtual Teaching Resources Based on Digital Three‐Dimensional Technology
CN111125541A (en) Method for acquiring sustainable multi-cloud service combination for multiple users
Rastogi et al. Unsupervised Classification of Mixed Data Type of Attributes Using Genetic Algorithm (Numeric, Categorical, Ordinal, Binary, Ratio-Scaled)
Zhang et al. Formation and Characteristics of Multi-Dimensional Tibetan Traditional Motion from the Perspective of Data Mining
Anari et al. Optimizing membership functions using learning automata for fuzzy association rule mining

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