CN112417171A - Data augmentation method for knowledge graph representation learning - Google Patents
Data augmentation method for knowledge graph representation learning Download PDFInfo
- Publication number
- CN112417171A CN112417171A CN202011319011.8A CN202011319011A CN112417171A CN 112417171 A CN112417171 A CN 112417171A CN 202011319011 A CN202011319011 A CN 202011319011A CN 112417171 A CN112417171 A CN 112417171A
- Authority
- CN
- China
- Prior art keywords
- triples
- knowledge graph
- candidate
- rule
- knowledge
- 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
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000013434 data augmentation Methods 0.000 title claims abstract description 23
- 230000003190 augmentative effect Effects 0.000 claims abstract description 34
- 230000003416 augmentation Effects 0.000 claims abstract description 19
- 238000001914 filtration Methods 0.000 claims abstract description 7
- 230000001502 supplementing effect Effects 0.000 claims abstract description 4
- 238000005065 mining Methods 0.000 claims description 14
- 230000000694 effects Effects 0.000 abstract description 4
- 238000012549 training Methods 0.000 description 4
- 238000009795 derivation Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000001228 spectrum Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 239000007787 solid Substances 0.000 description 2
- 239000013589 supplement Substances 0.000 description 2
- 239000002585 base Substances 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 239000012458 free base Substances 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000001902 propagating effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a data augmentation method for knowledge graph representation learning, which comprises the following steps: given a knowledge-graph, triples in the current knowledge-graph are first treated as observed facts and logical rules are mined from them. Next, through a rule instantiation step, the potential fact containing partial noise is derived by using the currently observed fact, and the potential fact is used as a candidate triple for augmentation. Since the representation learning model has certain discrimination capability on the triples, the invention trains and obtains a knowledge graph embedded model based on the currently observed facts. And finally, verifying the candidate triples by combining the embedded model, filtering out noise, obtaining augmented triples and supplementing the augmented triples into the current knowledge graph. The above steps are repeated until the maximum number of iterations is reached. By applying the method and the system, the comprehensible data of the user can be enlarged for the knowledge graph, and the learning effect of the knowledge graph representation is improved.
Description
Technical Field
The invention relates to the technical field of knowledge graphs, in particular to a data augmentation method for knowledge graph representation learning.
Background
In 2012, Google corporation first proposed a new concept, the knowledge graph, which improves the quality of the search by introducing the knowledge graph to structure information about the search targets. From the content perspective, the knowledge graph is mainly composed of interconnected entities and their attributes; and in essence, it can be viewed as a knowledge base built based on a semantic network, where each piece of knowledge can be represented by a triplet. For example (france, capital, paris), characterizes a piece of knowledge (fact) in the real world: the first in france is paris. Since many scenes in the real world are suitable for representation by knowledge graph, in recent years, the construction and application work on knowledge graphs has become a new research hotspot. Currently, a large set of quality knowledge maps are emerging in the industry, such as Freebase, which is widely used in real world applications. Since knowledge-graphs are symbol-based structured data that is not readily applied directly to other tasks, how to handle such symbolic representations becomes a significant challenge.
This research problem is called representation learning of the knowledge graph (or embedding of the knowledge graph). The goal of knowledge graph representation learning is to hopefully assist downstream tasks in the real world by vectorizing elements throughout the knowledge graph. The problem is modeled as a mapping from symbolic representation to vector representation, while requiring that a new representation in the vector space implies as much structural information as possible in the original symbolic space. Research has shown that representation learning this process typically relies on a large number of triples to capture potential semantic patterns. However, the knowledge-graph is naturally incomplete, and its intrinsic knowledge is usually limited, which results in the model-derived embedded representation often being poorly expressible. Intuitively, data is augmented for the knowledge graph spectrum, data information required by learning is represented by the knowledge graph spectrum, and the strategy is a feasible solution.
In the field of computer vision, data augmentation has proven to be a simple, effective technique and is widely used. Due to the characteristics of the picture, the modes of translation, rotation, scaling and the like are very natural data augmentation modes. However, in the knowledge-graph, the meaning of the three-unit expression may be changed greatly when a piece of structured knowledge is changed a little bit, which results in that the data augmentation operation in the knowledge-graph field is not trivial. Note that in a knowledge graph, logical rules are generally used for knowledge reasoning, and the logical rules themselves imply some generalized background information, which makes the data augmentation using the logical rules a feasible solution.
In the above process, one key of data augmentation using logic rules is how to ensure the quality of augmented data. Although logic rules are often accurate, it is still difficult to guarantee the quality of the new knowledge they infer. Filtering by manual means is costly, as the augmented data is difficult for users to discern. Based on this, it is necessary to design an automated processing method to ensure the quality of the augmented data. Meanwhile, because it is difficult to ensure the complete accuracy of the augmented data, a potential mechanism is needed in the process to ensure that the augmented data is understandable for the user, so that the whole augmented process is user-controllable.
The existing methods are of little concern for data augmentation of knowledge spectra, and although some methods generate new triples through various strategies, the generation process is not understandable to the user, and the quality of generating triples is not carefully considered.
Disclosure of Invention
The purpose of the invention is as follows: in view of the problems and deficiencies of the prior art, the invention aims to provide a data augmentation method for knowledge graph representation learning, which can continuously supplement high-quality augmentation triples understandable by users for knowledge graphs by utilizing rich background information contained in logic rules. The invention can enrich the original knowledge map information, ensure the quality of the augmentation data, improve the learning effect and provide the visual explanation of the augmentation data for the user.
The technical scheme is as follows: in order to achieve the above object, the technical solution adopted by the present invention is a data augmentation method for knowledge graph representation learning, comprising the following steps:
(1.1) for the current given knowledge graph, mining a logic rule from the knowledge graph by using an association rule mining algorithm;
(1.2) deducing by combining facts observed in a current knowledge graph based on the logic rules obtained in the step (1.1), and instantiating the logic rules to obtain candidate triples to be augmented;
(1.3) for the currently given knowledge graph, using a representation learning model to obtain an embedded representation of an entity and a relation;
(1.4) based on the step (1.2) and the step (1.3), verifying the candidate triples by using the embedded representation of the knowledge graph, filtering out noise from the candidate triples, and supplementing the filtered candidate triples into the current knowledge graph as augmented triples;
(1.5) repeating said steps (1.1) to (1.4) until a maximum number of iterations is reached.
Further, the step (1.2) comprises the steps of:
(2.1) traversing each mined logic rule, instantiating an entity variable in a rule body by using a specific entity aiming at the rule body in the logic rule, simultaneously limiting each triple in the instantiated rule body to be from a currently given knowledge graph, and after the rule body is instantiated, instantiating a rule head so as to obtain the instantiated logic rule;
and (2.2) taking a rule head in the instantiated logic rule as a candidate triple to be augmented, and taking the instantiated logic rule as an explanation of knowledge graph augmentation.
Further, the step (1.4) comprises the steps of:
(3.1) scoring all candidate triples by using an embedded model of the current knowledge graph, wherein the score represents the probability that the candidate triples are true;
(3.2) setting a threshold value, and marking the candidate triple with the score smaller than the threshold value as a counterexample;
(3.3) establishing propagation influence structures for the candidate triples by using the logic rules, wherein each propagation influence structure corresponds to a specific scoring updating formula;
(3.4) searching a propagation influence structure established between the candidate triples and the rest of the candidate triples by using the candidate triples marked as counterexamples, and scoring and updating the rest of the candidate triples according to a corresponding scoring updating formula;
and (3.5) selecting the candidate triples with the scores larger than or equal to the threshold value from the score-corrected candidate triples as augmented triples.
Has the advantages that: (1) the method has the advantages that automatic data augmentation is carried out on the knowledge graph, external information does not need to be additionally introduced, the quality of augmented data is guaranteed, and meanwhile the ability of representing learning of the knowledge graph is improved. (2) The whole augmentation process is understandable to the user, and in the whole augmentation process, not only is an augmentation triple provided, but also a basis for augmenting the triple is provided to serve as an explanation for the user, so that the user can conveniently check the augmentation process, and the whole augmentation process is controllable.
Drawings
FIG. 1 is an overall flow diagram of the present invention;
fig. 2 is a schematic diagram of the structure of propagation influence between triplets designed by the present invention.
Detailed Description
The present invention is further illustrated by the following figures and specific examples, which are to be understood as illustrative only and not as limiting the scope of the invention, which is to be given the full breadth of the appended claims and any and all equivalent modifications thereof which may occur to those skilled in the art upon reading the present specification.
The invention utilizes the logic rule to deduce the potential knowledge as the candidate triple, and the reasoning process is used as the explanation to the user. And (3) scoring the candidate triples through an embedded model, establishing a propagation influence mechanism for the triples by using a logic rule, automatically labeling counter-example triples in the candidate set, and propagating the labeled influence to other triples so as to adjust the distribution condition of scoring of the whole candidate set. And selecting potential positive examples as augmented triples to supplement according to the final score distribution. The whole process is iterated continuously, so that the data augmentation module and the representation learning module are mutually promoted.
The overall process of the invention is shown in fig. 1, and comprises 5 parts: the method comprises the steps of mining Hohn rules (rules for short) according to triple information in a current knowledge graph by using an association rule mining algorithm, deriving candidate triple sets for augmentation according to the rules, training in the current knowledge graph by using a representation learning model to obtain embedded representations of entities and relations, filtering the candidate triples by using the embedded representations to obtain augmented triples, supplementing the augmented triples into the knowledge graph, and repeating the steps to enable a data augmentation module and a representation learning module to promote each other.
The specific embodiments are described below:
1. mining rules by using association rule mining algorithm according to triple information in current knowledge graph
For a given knowledge graph, the contained triplets are taken as known knowledge. And mining by using a mainstream association rule mining algorithm AMIE. The mined rule has a confidence level indicating the accuracy of the rule. According to the knowledge graph with different scales, a user can set the corresponding rule confidence threshold and limit the length of the rule body, so that the quality of the augmentation data and the time overhead of rule mining are ensured to be within a reasonable range, for example, the rule confidence threshold is set to be 0.8, and the length of the rule body is limited to be 2.
2. Deriving a set of candidate triples for augmentation according to rules
The method for generating the candidate triple for augmentation by using the rule mainly comprises the following two sub-steps:
(1) because the mined rules contain entity variables, the invention instantiates the variables through specific entities, thereby obtaining the instantiated rules. In the instantiation stage, the rule body can only be instantiated by using the currently observed triple, and after the rule body is instantiated, the rule head is also instantiated. In order to ensure that the derived triple is unknown, it is necessary to ensure that the instantiated rule header does not appear in the currently observed triple, thereby ensuring the amount of information that the augmented data has;
(2) after the instantiated rules are obtained, the instantiated results can be viewed as a forward reasoning process. The instantiated rule body is known knowledge, and the instantiated rule head is derived unknown knowledge. Thus, instantiated rules may be relied upon and interpreted as an augmentation of the triplet. And finally, collecting all rule heads of the instantiated rules, and combining the rule heads into a candidate triple set with augmented data.
3. Training in current knowledge map by using representation learning model to obtain embedded representation of entity and relation
The invention uses a main flow knowledge graph to represent a learning model, namely TransE, to train the current knowledge graph to obtain a trained embedded model, so that the embedded model has certain prediction capability on unknown triples. Because the whole processing flow is an iterative processing mode, the incremental learning method adopts an incremental learning process and achieves the effect of incremental learning by setting initialization parameters. The concrete measures are as follows:
(1) during the first training, performing random initialization on the model parameters by adopting a conventional mode;
(2) in the later iteration process, the current model is initialized by using the model parameters learned in the previous round each time. By a better initial value, the model can converge to a better result.
4. Filtering the candidate triples by using the embedded expression to obtain augmented triples which are supplemented into the knowledge graph
Scoring the candidate triples by using the embedded model obtained by current training, so as to obtain an initial score distribution on the candidate triples, wherein the score represents the possibility that the triples are positive examples;
in order to ensure the quality of augmented triples, a threshold value is set, for example 0.5, triples with initial scores less than the threshold value are automatically marked as counter-examples, and these triples can be directly filtered and certainly not be supplemented into the knowledge-graph as augmented data. And the candidate triples with scores greater than or equal to the threshold are taken as potential positive examples, which need subsequent further filtering;
rules are used to build propagation impact structures for between candidate triples. The invention is mainly concerned with ensuring the quality of the augmented data, i.e. with the desire to reduce candidates as much as possibleThe scores of counterexamples are concentrated and therefore are designed primarily for the case labeled as counterexamples. As shown in FIG. 2, the present invention mainly designs three propagation structures, where t represents a triple and μ represents a rule. The core idea is to utilize the equivalence of the inverse proposition and the original proposition. The rule establishes a relation between candidate triples, and by regarding a triplet as a proposition, a forward reasoning process can be regarded as an original proposition, namely, one triplet is correct, and by the rule, another triplet can be deduced to be also correct. The contrary-negative proposition is that given that a triple is wrong, then it can be deduced from the rule that another triple may also be wrong. In FIG. 2, the solid circles represent triplets (t) labeled as counter-examples1、t3And t6) Solid open circles indicate the triads (t) that need to be affected2、t4、t5And t8) The dotted open circles represent the currently observed triplets (t)7). The first propagation structure is the most basic negative proposition idea; the second propagation structure means that a conclusion needs to be derived from two triples together in an original proposition, and then in a negative proposition, one of the two original triples may be wrong; the third propagation structure means that two original propositions are known, and one derivation is wrong, and the rule confidence of the derivation process is higher than that of the other original proposition, so that the derivation process of the other original proposition can be suspected to be wrong. And finally, establishing a propagation relation among the candidate triples by using rules through the three propagation structures. Meanwhile, corresponding updating formulas need to be designed according to the characteristics of the three propagation structures. The invention is mainly based on two intuitive recognitions: if the confidence coefficient of the rule is higher, the reasoning process of the inverse negative proposition is more accurate, and the influence on the propagation is more severe; the greater the probability that the current counterexample is false, the more strongly the propagation will be affected. The following are four specific update formulas corresponding to the three propagation structures described above:
p(t2)＝min(0,p(t2)-(1-p(t1))·λ1),
p(t8)＝min(0,p(t8)-(1-p(t6))·λ4),
wherein p (t) represents the score of the triplet, λ represents the confidence of the corresponding rule μ;
and traversing each labeled counterexample according to the propagation influence structure among the candidate triples, searching the propagation influence structure established between the candidate triples and the rest triples, and scoring and updating the rest triples according to the corresponding scoring updating formula. Meanwhile, in order to further expand the propagation influence, if the currently influenced triple is also a counterexample, the scoring of the other triples is continuously updated by using the triple;
and after the score of the whole candidate triple is corrected, the triple with the final score being more than or equal to the threshold value of 0.5 is taken as augmentation data and is supplemented into the current knowledge graph.
5. Repeating the above parts makes the data augmentation module and the representation learning module mutually promote
The above parts together constitute a round of iteration. By repeating the above parts for multiple iterations, the data augmentation module and the representation learning module can be facilitated. To reduce the time overhead caused by multiple iterations, 2 is chosen as the maximum number of iterations. On one hand, as the data of multiple rounds is expanded, the rule mining part can capture rich semantic patterns and mine a large number of reliable rules. Table 1 shows the case of rule mining and data augmentation in the present invention on two widely used reference data sets. The number of augmented triplets may account for 4% to 30% of the original data on both data sets.
Table 1: rule mining and data augmentation in the present invention
| Data set | Number of rules mined | Original triple number | Augmenting triple numbers |
| WN18 | 41 | 141442 | 6667 |
| FB15K | 896 | 483142 | 152760 |
On the other hand, with multiple rounds of iteration, the knowledge graph is greatly enriched, and the embedded model can capture the relation between the entity and the relation in a more explicit mode, so that a stronger embedded representation is obtained. The model is tested by using the link prediction task on the two data sets, and the test result shown in the table 2 can obviously improve the effect of originally representing the learning model after multiple rounds of data augmentation. For example, in WN18, each index is raised by nearly 0.01 compared to the original TransE model, and in the more densely related data set FB15K, it can be raised by 0.04 to 0.07.
Table 2: the invention promotes the learning model of knowledge graph representation
Meanwhile, the invention can provide the explanation of the augmented triple for the user as the basis for augmentation. Table 3 shows some illustrative examples of augmented triplets. For example, one of the augmented triplets (Johnny Depp, national, u.s.a.) is based on Johnny Depp's place of origin being Kentucky, which is a state in the united states, so Johnny Depp's nationality may be in the united states. Such an interpretation may facilitate user verification, making the entire augmentation process controllable.
Table 3: example of the invention providing augmented triple interpretation for a user
Claims (3)
1. A data augmentation method for knowledge graph representation learning is characterized by comprising the following steps:
(1.1) for the current given knowledge graph, mining a logic rule from the knowledge graph by using an association rule mining algorithm;
(1.2) deducing by combining facts observed in a current knowledge graph based on the logic rules obtained in the step (1.1), and instantiating the logic rules to obtain candidate triples to be augmented;
(1.3) for the currently given knowledge graph, using a representation learning model to obtain an embedded representation of an entity and a relation;
(1.4) based on the step (1.2) and the step (1.3), verifying the candidate triples by using the embedded representation of the knowledge graph, filtering out noise from the candidate triples, and supplementing the filtered candidate triples into the current knowledge graph as augmented triples;
(1.5) repeating said steps (1.1) to (1.4) until a maximum number of iterations is reached.
2. The knowledge-graph-representation-learning-oriented data augmentation method of claim 1, wherein the step (1.2) comprises the steps of:
(2.1) traversing each mined logic rule, instantiating an entity variable in a rule body by using a specific entity aiming at the rule body in the logic rule, simultaneously limiting each triple in the instantiated rule body to be from a currently given knowledge graph, and after the rule body is instantiated, instantiating a rule head so as to obtain the instantiated logic rule;
and (2.2) taking a rule head in the instantiated logic rule as a candidate triple to be augmented, and taking the instantiated logic rule as an explanation of knowledge graph augmentation.
3. The knowledge-graph-representation-learning-oriented data augmentation method of claim 2, wherein the step (1.4) comprises the steps of:
(3.1) scoring all candidate triples by using an embedded model of the current knowledge graph, wherein the score represents the probability that the candidate triples are true;
(3.2) setting a threshold value, and marking the candidate triple with the score smaller than the threshold value as a counterexample;
(3.3) establishing propagation influence structures for the candidate triples by using the logic rules, wherein each propagation influence structure corresponds to a specific scoring updating formula;
(3.4) searching a propagation influence structure established between the candidate triples and the rest of the candidate triples by using the candidate triples marked as counterexamples, and scoring and updating the rest of the candidate triples according to a corresponding scoring updating formula;
and (3.5) selecting the candidate triples with the scores larger than or equal to the threshold value from the score-corrected candidate triples as augmented triples.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011319011.8A CN112417171B (en) | 2020-11-23 | 2020-11-23 | Knowledge graph representation learning-oriented data augmentation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011319011.8A CN112417171B (en) | 2020-11-23 | 2020-11-23 | Knowledge graph representation learning-oriented data augmentation method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112417171A true CN112417171A (en) | 2021-02-26 |
| CN112417171B CN112417171B (en) | 2023-10-03 |
Family
ID=74778440
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011319011.8A Active CN112417171B (en) | 2020-11-23 | 2020-11-23 | Knowledge graph representation learning-oriented data augmentation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112417171B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113626612A (en) * | 2021-08-13 | 2021-11-09 | 第四范式（北京）技术有限公司 | Prediction method and system based on knowledge graph reasoning |
| CN113722611A (en) * | 2021-08-23 | 2021-11-30 | 讯飞智元信息科技有限公司 | Method, device and equipment for recommending government affair service and computer readable storage medium |
| CN114741460A (en) * | 2022-06-10 | 2022-07-12 | 山东大学 | Knowledge graph data expansion method and system based on association between rules |
| CN116955653A (en) * | 2023-09-20 | 2023-10-27 | 首都信息发展股份有限公司 | Knowledge graph data enhancement method and device and electronic equipment |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3156949A2 (en) * | 2015-10-16 | 2017-04-19 | Baidu USA LLC | Systems and methods for human inspired simple question answering (hisqa) |
| CN108304933A (en) * | 2018-01-29 | 2018-07-20 | 北京师范大学 | A kind of complementing method and complementing device of knowledge base |
| CN109614495A (en) * | 2018-08-08 | 2019-04-12 | 广州初星科技有限公司 | A kind of associated companies method for digging of combination knowledge mapping and text information |
| CN109948150A (en) * | 2019-03-01 | 2019-06-28 | 北京航空航天大学 | The high performance service context of knowledge based map finds method in a kind of multi-domain environment |
| CN111144570A (en) * | 2019-12-27 | 2020-05-12 | 福州大学 | Knowledge representation method combining logic rules and confidence degrees |
| CN111191460A (en) * | 2019-12-30 | 2020-05-22 | 福州大学 | Relation prediction method combining logic rule and fragmentation knowledge |
| CN111783902A (en) * | 2020-07-30 | 2020-10-16 | 腾讯科技（深圳）有限公司 | Data augmentation and service processing method and device, computer equipment and storage medium |
-
2020
- 2020-11-23 CN CN202011319011.8A patent/CN112417171B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3156949A2 (en) * | 2015-10-16 | 2017-04-19 | Baidu USA LLC | Systems and methods for human inspired simple question answering (hisqa) |
| CN108304933A (en) * | 2018-01-29 | 2018-07-20 | 北京师范大学 | A kind of complementing method and complementing device of knowledge base |
| CN109614495A (en) * | 2018-08-08 | 2019-04-12 | 广州初星科技有限公司 | A kind of associated companies method for digging of combination knowledge mapping and text information |
| CN109948150A (en) * | 2019-03-01 | 2019-06-28 | 北京航空航天大学 | The high performance service context of knowledge based map finds method in a kind of multi-domain environment |
| CN111144570A (en) * | 2019-12-27 | 2020-05-12 | 福州大学 | Knowledge representation method combining logic rules and confidence degrees |
| CN111191460A (en) * | 2019-12-30 | 2020-05-22 | 福州大学 | Relation prediction method combining logic rule and fragmentation knowledge |
| CN111783902A (en) * | 2020-07-30 | 2020-10-16 | 腾讯科技（深圳）有限公司 | Data augmentation and service processing method and device, computer equipment and storage medium |
Non-Patent Citations (4)
| Title |
|---|
| LI GUANG YAO等: "rule-based data augmentation for knowledge graph embedding", AI OPEN, vol. 2, pages 186 - 196, XP093055466, DOI: 10.1016/j.aiopen.2021.09.003 * |
| 刘绍南;杨鸿波;侯霞;: "文物知识图谱的构建与应用探讨", 中国博物馆, no. 04, pages 118 - 125 * |
| 谭晓;张志强;: "知识图谱研究进展及其前沿主题分析", 图书与情报, no. 02, pages 50 - 63 * |
| 阳小华;张硕望;欧阳纯萍;: "中文关系抽取技术研究", 南华大学学报(自然科学版), vol. 32, no. 01, pages 66 - 72 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113626612A (en) * | 2021-08-13 | 2021-11-09 | 第四范式（北京）技术有限公司 | Prediction method and system based on knowledge graph reasoning |
| CN113722611A (en) * | 2021-08-23 | 2021-11-30 | 讯飞智元信息科技有限公司 | Method, device and equipment for recommending government affair service and computer readable storage medium |
| CN113722611B (en) * | 2021-08-23 | 2024-03-08 | 讯飞智元信息科技有限公司 | Recommendation method, device and equipment for government affair service and computer readable storage medium |
| CN114741460A (en) * | 2022-06-10 | 2022-07-12 | 山东大学 | Knowledge graph data expansion method and system based on association between rules |
| CN114741460B (en) * | 2022-06-10 | 2022-09-30 | 山东大学 | Knowledge graph data expansion method and system based on association between rules |
| CN116955653A (en) * | 2023-09-20 | 2023-10-27 | 首都信息发展股份有限公司 | Knowledge graph data enhancement method and device and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112417171B (en) | 2023-10-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112417171A (en) | Data augmentation method for knowledge graph representation learning | |
| CN109145939B (en) | Semantic segmentation method for small-target sensitive dual-channel convolutional neural network | |
| CN108256065B (en) | Knowledge graph reasoning method based on relation detection and reinforcement learning | |
| CN109784149B (en) | Method and system for detecting key points of human skeleton | |
| CN111160467A (en) | Image description method based on conditional random field and internal semantic attention | |
| CN114780691B (en) | Model pre-training and natural language processing method, device, equipment and storage medium | |
| CN114791958B (en) | Zero sample cross-modal retrieval method based on variational self-encoder | |
| CN113239131A (en) | Low-sample knowledge graph completion method based on meta-learning | |
| CN111598252B (en) | University computer basic knowledge problem solving method based on deep learning | |
| Cohen et al. | A provably correct learning algorithm for latent-variable PCFGs | |
| CN115964459B (en) | Multi-hop reasoning question-answering method and system based on food safety cognition spectrum | |
| CN115269861A (en) | Reinforced learning knowledge graph reasoning method based on generative confrontation and imitation learning | |
| CN113033410B (en) | Domain generalization pedestrian re-recognition method, system and medium based on automatic data enhancement | |
| CN110765781A (en) | Man-machine collaborative construction method for domain term semantic knowledge base | |
| CN114399661A (en) | Instance awareness backbone network training method | |
| CN111914772B (en) | Age identification method, age identification model training method and device | |
| CN113032582A (en) | Knowledge graph based entity unified model establishment and entity unified method | |
| CN103150329A (en) | Word alignment method and device of bitext | |
| CN105787505A (en) | Infrared image clustering segmentation method combining sparse coding and spatial constraints | |
| CN116737956A (en) | Entity alignment method and device for multi-mode knowledge graph | |
| CN116701647A (en) | Knowledge graph completion method and device based on fusion of embedded vector and transfer learning | |
| CN111144134A (en) | Translation engine automatic evaluation system based on OpenKiwi | |
| CN115035304A (en) | Image description generation method and system based on course learning | |
| Xu et al. | Bayesian network structure learning using causality | |
| CN109409226A (en) | A kind of finger vena plot quality appraisal procedure and its device based on cascade optimization CNN |
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 |