CN109033160A - A kind of knowledge mapping dynamic updating method - Google Patents
A kind of knowledge mapping dynamic updating method Download PDFInfo
- Publication number
- CN109033160A CN109033160A CN201810627957.7A CN201810627957A CN109033160A CN 109033160 A CN109033160 A CN 109033160A CN 201810627957 A CN201810627957 A CN 201810627957A CN 109033160 A CN109033160 A CN 109033160A
- Authority
- CN
- China
- Prior art keywords
- entity
- updated
- extension
- knowledge mapping
- entities
- 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.)
- Pending
Links
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of knowledge mapping dynamic updating methods, for solving the stationary problem between encyclopaedic knowledge map and its data source.The present invention is using the Hot Contents on WWW as starting point；Name entity is therefrom extracted as kind of a fructification, these usual entities are the entities that possible update.Then crawl other entities relevant with kind fructification are used as extension entity on encyclopaedia website.Then, a certain number of entities are obtained from encyclopaedia website and carry out Feature Engineering, and excavate the more new information of substance feature characterization using machine learning algorithm, construct fallout predictor；The high entity of update probability is filtered out from extension entity using fallout predictor.Finally, in the case where the limitation of data source access number, realizing that the dynamic of knowledge mapping updates using the high extension entity of kind of fructification and update probability as the object updated.
Description
Technical field
The invention belongs to knowledge mapping fields, are related to a kind of encyclopaedic knowledge map dynamic updating method.
Background technique
Knowledge mapping can be constructed by various forms of data sources, and wherein encyclopaedic knowledge map is exactly to pass through encyclopaedia
What the data on website were built.Usually, data the managing and maintaining due to specification on encyclopaedia website, data cover
Wide, content is accurate, and the knowledge mapping built also has better quality.But due to the knowledge and information on encyclopaedia website
Variation, already existing data can face out-of-date problem in encyclopaedic knowledge map；In addition, for occurring on encyclopaedia website
Completely new vocabulary, knowledge mapping can not timely include them.
In order to guarantee the not out-of-date update it is necessary to keep data and data source in knowledge mapping of the data in knowledge mapping
It is synchronous as far as possible.A kind of basic method is the data by obtaining entire data source, to rebuild knowledge mapping, this
It is the way that current most of encyclopaedic knowledge maps are updated.Certain encyclopaedia websites, such as wikipedia can periodically provide complete
The data set and data change information stood, provide a great convenience the acquisition of data.But also some encyclopaedia websites are simultaneously
Such data set, such as Baidupedia, interaction encyclopaedia will not be provided；If it is desired to updating to encyclopaedia data, crawler can only be passed through
Crawl some or all of data.Do so some drawbacks: first, although wikipedia provides the variation letter of data set
Breath, but these data sets are usually to be provided using the moon as the period；In this way, knowledge mapping still cannot timely keep up with encyclopaedia enough
The information change of website.Second, for not providing the encyclopaedia website of data set downloading, if grabbing the number of whole station by crawler
According to for the entity of ten million number of levels, machine runs without interruption one month and also different surely crawled.No matter third,
It is downloaded either with or without data set, obtains whole data sets all very greatly, this can expend very more bandwidth；And encyclopaedia website,
Also it can limit frequently access is crawled, so that the acquisition of new data is highly difficult.So finding a kind of knowledge mapping dynamic
Update method, this method keep the timeliness of knowledge mapping data under conditions of avoiding bandwidth resources from wasting, and have very big
Research significance.
Summary of the invention
Technical problem: the present invention provides a kind of knowledge mapping dynamic updating method, and this method avoid network bandwidths and meter
The waste for calculating resource, greatly reduces the time lag of data in knowledge mapping, can it is automatic, frequently, efficiently to knowledge graph
Spectrum is updated, and knowledge mapping may be implemented, and frequently efficient more new task is also protected while ensure that lesser amount of access
The quality for updating work is demonstrate,proved.
Technical solution: the present invention searches vocabulary as starting point using the heat of popular title and search engine on social network sites, from
These titles and heat, which are searched, excavates entity word in word, as kind of a fructification, these usual entities are the entities that possible update.
Then related entities are grabbed on encyclopaedia website to be extended.Then, by carrying out Feature Engineering in certain amount entity information,
Based on these features, the fallout predictor that can predict future update situation is trained with machine learning method.Finally, just with seed
Entity and the higher extension entity of update probability are realized in the case where the limitation of data source access number as the object updated
The dynamic of knowledge mapping updates.
Knowledge mapping dynamic updating method of the invention, includes the following steps:
1) vocabulary is searched as starting point, from these titles and heat using the popular title on social network sites and the heat on search engine
It searches and extracts entity in word, using these entities as kind fructification to be updated；
2) internal chaining in the Abstract and Infobox on the encyclopaedia page of crawl kind fructification, as extension entity；
3) entity word is grabbed from encyclopaedia website, therefrom extracts feature, with history renewal frequency for Label value, building
Training dataset establishes regression model using machine learning method, trains fallout predictor from the training dataset of building；
4) using the renewal frequency of fallout predictor prediction extension entity, forecast updating frequency higher preceding K are picked out
Entity is extended, K is that extension entity updates the upper limit of the number value；
5) the extension entity that kind of fructification and the step 4) are picked out carries out more knowledge mapping as upgating object
Newly.
In the preferred embodiment of dynamic updating method of the present invention, in the step 1), entity is extracted as follows:
Name entity 1-a) is extracted first using name Entity recognition to the title grabbed；
1-b) then utilize participle technique, from title obtain identification less than candidate entity word list；
Part-of-speech tagging 1-c) is carried out to candidate entity word, weeding out auxiliary word, number, these do not have the candidate vocabulary of practical significance,
Then verify whether these candidate words are entity words on encyclopaedia website, using entity word and the name entity of extraction as seed reality
Body.
In the preferred embodiment of dynamic updating method of the present invention, training data is constructed according to following characteristics in the step 3)
Collection:
Number of weeks existing for 3-1): entity is present in the time on encyclopaedia website, and new entity is more likely to be updated；
3-2) update times: the number that entity is updated in total, the number being updated in the past is more, it is following more may by after
It is continuous to update；
3-3) browsing time: the number that an entity is seen is more, illustrates more welcome, is also more likely to be updated；
3-4) number of links: easier by it when the hyperlink that a physical page includes other resources or entity is more
His resource or entity influence；
3-5) internal chaining number: if being linked to other entities, the change of other entities may also will affect oneself；
3-6) content of pages length: an article is longer, and content is abundanter, easier to be updated；
3-7) clip Text length: abstract contains the description most succinct to entity, makes a summary longer, information content is bigger；
3-8) average all renewal frequency: past updates more frequent, then future may also update it is frequent.
In the preferred embodiment of dynamic updating method of the present invention, according to following specific steps to knowledge mapping in the step 5)
It is updated:
Firstly, more content of the new seed entity in knowledge mapping, extension entity is then updated according to priority and is being known
Know the content in map: for the entity being originally not present in knowledge mapping, being inserted into knowledge mapping with highest priority；It is right
The entity existing for those scripts provides forecast updating frequency first with fallout predictor, then preferentially to upper subsynchronous expectation later
The maximum extension entity of renewal frequency is updated, until all extension entities have all been updated completion or extension entity update
The extension entity that quantity reaches setting updates the upper limit of the number value K, and wherein K is determined according to network bandwidth and computing resource.
In the preferred embodiment of dynamic updating method of the present invention, priority when extension entity updates is calculate by the following formula excellent
First grade value determines that priority value is bigger, then priority is higher:
E [u (x)]=P (x) × (tnow-ts(x))
Wherein, E [u (x)] be to the priority value that is updated of extension entity, u (x) be it is upper it is subsynchronous after update when
Between, P (x) is the renewal frequency predicted by fallout predictor, tnowFor current time, tsIt (x) is the time updated entity x last time, when
When x is novel entities, ts(x)=- ∞.
Encyclopaedic knowledge map dynamic updating method of the invention does not need periodically to obtain whole numbers on encyclopaedia website
According to and frequently knowledge mapping can be updated with fraction of amount of access, avoid existing method well to network bandwidth
Waste, also greatly reduce the time lag of data in knowledge mapping.
The present invention finds the entity that may have occurred update using enlightening rule；And by selecting feature, excavate
Relationship between entity and renewal frequency out, constructs fallout predictor, further finds the entity that may be had updated.Realize energy
Enough automatic, frequently, the dynamic updating method being efficiently updated to knowledge mapping reduces in knowledge mapping renewal process not
Necessary update.Problem is expressed as follows:
Wherein, tnIt (x) is the last time renewal time of entity x on encyclopaedia website, tsIt (x) is to synchronize the entity x last time
Time is (if x is novel entities, ts(x)=- ∞), the intention of formula expression are as follows: be up to K's in the number of processing entities
Under the conditions of, after subsynchronous in searching as much as possible, the entity that is updated on encyclopaedia website.
The utility model has the advantages that compared with prior art, the present invention having the advantage that
Compared to the update mode of current most of knowledge mappings, the present invention is efficiently realized to knowledge mapping part more
Newly, it updates those reality and is more likely to the entity changed on encyclopaedia website.When existing encyclopaedic knowledge map updates, mostly
Need to obtain data whole in nearest data source；No matter whether encyclopaedia website provides data set downloading, this can all be wasted largely
Network bandwidth.In the present invention, it is updated every time from network hotspot, extracts kind of a fructification, then sought on encyclopaedia website
It looks for related entities to be extended, finally the entity picked out is updated；Demand very little of the whole process to Internet resources, section
Network bandwidth is saved.It is based on the judgement of network hotspot and some heuristic rules and fallout predictor simultaneously, so that choosing update
Entity quality has guarantee, is updated to those entities with greater need for update.In addition, due to the data volume to access every time
It is very small, it may be implemented frequently to be updated the data in knowledge mapping, greatly reduce the lag of knowledge mapping data
Property.
It is proved by experimental analysis, using knowledge mapping dynamic updating method proposed by the present invention, knowledge graph may be implemented
The frequent efficient more new task of spectrum.Based on it is didactic rule and sequence of the fallout predictor to renewal frequency so that those with greater need for
The entity of update is selected out, while ensure that lesser amount of access, also ensures the quality for updating work.The present invention
In, plant the validity of fructification, the validity of the extension entity that fallout predictor is picked out is in accuracy rate, recall rate, F1 value, and
The outstanding effect of the update method is all sufficiently demonstrated by the indexs such as MAP, nDCG, AUC.
Detailed description of the invention
Fig. 1 is general frame schematic diagram of the invention；
Fig. 2 is the flow diagram of more new algorithm in the present invention.
Specific embodiment
With reference to embodiments and Figure of description, the implementation process that the present invention will be described in detail.
The present invention is the table entity link method based on multiple knowledge base, including following 5 steps:
1) vocabulary is searched as starting point using the heat of popular title and search engine on social network sites, is searched from these titles and heat
Entity is extracted in word, using these entities as kind fructification to be updated.This process is the discovery of kind of fructification, is walked in detail
It is rapid as follows:
(1) hot spot title is grabbed, heat searches vocabulary
Because of the hot topic quality with higher of the real-time hot spot and Baidu's discussion bar of search dog search, in the present embodiment with
They are as the source for planting fructification；The hot spot of search dog search and the topic numbers of Baidu's discussion bar be all it is constant, the two is shared
50 titles.In addition, Baidupedia homepage, which can also provide heat, searches entry, these can directly be grabbed as seed entity.
(2) kind of a fructification is extracted from title
Entity recognition is named to title first, and has selected following classification as the entity class retained: '
PERSON ', ' LOCATION ', ' ORGANIZATION ', ' MISC ', ' CITY ', ' STATE_OR_PROVINCE ', '
COUNTRY’.’RELIGION’.’IDEOLOGY’。
Meanwhile also picking out some entity class and being abandoned, predominantly digital, the classifications such as time, itself difficulty has reality
Meaning: ' MONEY ', ' NUMBER ', ' ORDINAL ', ' PERCENT ', ' DATE ', ' TIME ', ' SET ', ' DURATION ' '
TITLE’。
In addition, carrying out part-of-speech tagging to the word in title, remain to clean subsequent word segmentation result.
Name Entity recognition, which is only used only, can omit many entities.In order to improve recall rate, participle technique is used, has been come
More candidate seed entities (word) are obtained, the effect for segmenting generation is as follows, can more comprehensively obtain entity candidate word.
| Search engine mode | Accurate model | |
| Total eclipse of the moon will appear the sky | Total eclipse/total eclipse of the moon/appears/sky | Total eclipse of the moon/appear/the sky |
| Easy thousand imperial or royal seal achievement of melt exposure | Easily/melt/thousand imperial or royal seals/achievement/exposure | Thousand imperial or royal seals of easy melt/achievement/exposure |
| Intend closed management in the Ming Tombs | 13/tri- mounds/Ming Tombs/quasi-/closing/management | The Ming Tombs/quasi-/closing/management |
Although participle increases candidate entity word, also band is come in many meaningless words.By choosing some meaningless words
Property (for example, auxiliary word, number):
' AD ', ' AS ', ' BA ', ' CC ', ' CD ', ' CS ', ' DEC ', ' DEG ', ' DER ', ' DEV ', ' DT ', ' ETC ', '
IJ ', ' JJ ', ' LB ', ' LC ', ' M ', ' OD ', ' ON ', ' P ', ' PN ', ' PU ', ' SB ', ' SP ', ' VA ', ' VC ', ' VE ', '
VV ', ' X '
Then, using part-of-speech tagging information, to be cleaned to word segmentation result.So far time as much as possible is had been obtained for
Entity word is selected, finally, they are directly retrieved on Baidupedia as entity word, it is whether genuine real to verify them
Body；If there is corresponding entry, then it is assumed that be entity word, be not otherwise just entity word.Kind fructification, which extracts, to be completed.
2) the related internal chaining on the encyclopaedia page of crawl kind fructification, as extension entity.
(1) extension of fructification is planted
The present invention passes through the internal chaining for selecting kind of fructification, to select and plant the relevant entity of fructification.Specifically, this reality
Apply the internal chaining for having selected to include in abstract and Infobox two parts in the kind fructification encyclopaedia page in example.It is interior in abstract
Link and kind fructification are more relevant, and Infobox contains the association attributes of kind of fructification.The physical correlation extended in this way is more
Height, and will not excessively be expanded too much because of internal chaining.
(2) extension entity has actually occurred the ratio updated.
The present embodiment acquires more days seed solid datas by this method, and is extended.After one month,
Collect these extension nearest month ratios updated of entity.The update ratio of available extension entity is on 30% left side
It is right.This result shows that:
It one, is effective by the method that related entities are found in abstract and Infobox internal chaining.
Two, extend in entity and only 30% updated in the recent period, directly synchronize in this way still suffer from 70% extension
Entity does not update, synchronizes to them, still results in waste.Fallout predictor is needed to screen and be more likely to update
Entity.
3) a certain number of entity words are grabbed from encyclopaedia, therefrom extract feature, are Label with history renewal frequency
Value constructs training dataset.Regression model is established using machine learning method for training data, as fallout predictor.
In the present embodiment, tens of thousands of physical pages are grabbed from Baidupedia, and therefrom extracted following feature:
Feature 1: existing number of weeks: entity is present in the time on encyclopaedia website, and new entity is more likely to be updated.
Feature 2: update times: the number that entity is updated in total, the number being updated in the past is more, the following more possible quilt
Continue to update.
Feature 3: browsing time: the number that an entity is seen is more, illustrates more welcome, is also more likely to be updated.
Feature 4: number of links: when the hyperlink that a physical page includes other resources or entity is more, easier quilt
Other resources or entity influence.
Feature 5: internal chaining number: if being linked to other entities, the change of other entities may also be will affect self.
Feature 6: content of pages length: an article is longer, and content is abundanter, easier to be updated.
Feature 7: clip Text length: abstract contains the description most succinct to entity, makes a summary longer, information content is bigger.
Feature 8: average week renewal frequency: past updates more frequent, then future may also update it is frequent.
In the present embodiment, the validity of test feature engineering by the following method:
The label value in training sample, class categories are changed into: specifically, what the last one month was updated, y (e)
> 0, as positive example；The last one moon, there is no update, y (e)=0, as counter-example.Then chi-square value (χ 2) and information are calculated
Gain (IG:Information Gain).Chi-square value can be used to correlation of the judging characteristic between class label, value
Smaller to illustrate that correlation is lower, feature is more useless.Information gain indicates to learn the information of feature X and make the letter of class label
Cease the degree of uncertainty reduction.Feature is more useful, and information gain value is bigger.The test result of the present embodiment meets expection and sets
Think.
Regression model is as prediction model using in machine learning, and the fallout predictor in the present embodiment has used ridge regression and random
Two kinds of models of forest, and using history week renewal frequency as forecast updating frequency, construct benchmark model.Ridge regression is in linear regression
On the basis of be added to regularization term, can reduce the risk of over-fitting.Random forest is carried out using decision tree as base learner
It is integrated, while the selection of random attribute is introduced in the training process, so that finally integrated Generalization Capability is because of individual study
Between diversity factor increase and promoted.
4) entity for being more likely to update is picked out from extension entity using fallout predictor.
In the present embodiment, the feature of extension entity is extracted, the renewal frequency of prediction is then provided using fallout predictor, and will more
Newly frequency is descending is ranked up, and K extension entity before being selected according to network bandwidth and computing resource situation, network bandwidth is got over
Greatly, computing resource is abundanter, and K is bigger, and the maximum value that K can be obtained is to extend the number of entity；Otherwise, network bandwidth is small, calculates
Resource is few, in order to guarantee update validity, K is set it is a little bit smaller, can take extension physical quantities half.
In the present embodiment, it tests first with Hold-out model to fallout predictor: training dataset is divided into two
Part, wherein 90% is used to train, 10% is used to test.It calculates on test set, all renewal frequencies predicted and actual week are more
MSE between new frequency.Then determine whether each test entity occurs after time stamp T by different threshold
It updates, precision-racall curve is obtained with this, and judge the robustness of regressor using AUC.
The present embodiment also assesses the effect of fallout predictor on true data set: judging that can fallout predictor be
The entity updated has been actually occurred, higher update probability is predicted.
In time t, grab popular title, then therefrom extract several fructification, by these kind of fructification into
Row extension, obtains extending entity accordingly；Then, it is predicted against extension entity, and according to the frequency of prediction height
It is sorted.Then, in t to t+30 days, one month time, the update status of these extension entities is collected.Most Zhongdao t
It until+30, checks in this month, how many extension entity is updated altogether.Later, it is updated for these
Extension entity judged, look at they whether be predicted device be discharged to front.The present embodiment has used following index to carry out
Assessment:
MAP: Average Accuracy, for assessing the effect of retrieval ranking.The effect of assessment prediction ranking is carried out in the present embodiment
Fruit, in the top to be considered to retrieve relevant, then having actually occurred, the entity ranking updated is more forward, and MAP value will be got over
It is high.
NDCG: common ranking evaluation index, for the accuracy for ranking of testing and assessing.In the present embodiment, fallout predictor returns
The ranking of entity, each entity correspond to a score value, these score values are exactly gain (Gain) value.These scores
It is added, is exactly CumulativeGain (storage gain).Those have actually occurred those of update entity, should be put into front.
Therefore, when being added these scores, each is needed divided by an incremental number, that is, loses value, and obtain DCG, finally
It is normalized.
Area below AUC:Precision-Recall curve, biggish AUC represent preferable effect.
Precision@n: n entity in the top has actually occurred the ratio updated.
Recall@n: having actually occurred in the entity updated, how much has been discharged to preceding n.
Fl@n: the harmonic-mean based on Precision and Recall, more attention smaller value.
In the multiple test performance of the present embodiment, in most cases Random Forest model is all showed best, therefore this
Invention is using Random Forest model as fallout predictor model.
5) using kind of fructification and the extension entity picked out as upgating object, knowledge mapping is updated.
Claims (5)
1. a kind of knowledge mapping dynamic updating method, which is characterized in that method includes the following steps:
1) vocabulary is searched as starting point using the popular title on social network sites and the heat on search engine, searches word from these titles and heat
In extract entity, using these entities as kind fructification to be updated；
2) internal chaining in the Abstract and Infobox on the encyclopaedia page of crawl kind fructification, as extension entity；
3) entity word is grabbed from encyclopaedia website, therefrom extracts feature, with history renewal frequency for Label value, building training
Data set establishes regression model using machine learning method, trains fallout predictor from the training dataset of building；
4) using the renewal frequency of fallout predictor prediction extension entity, the higher preceding K extension of forecast updating frequency is picked out
Entity, K are that extension entity updates the upper limit of the number value；
5) the extension entity that kind of fructification and the step 4) are picked out is updated knowledge mapping as upgating object.
2. knowledge mapping dynamic updating method according to claim 1, which is characterized in that in the step 1), according to such as
Under type extracts entity:
Name entity 1-a) is extracted first using name Entity recognition to the title grabbed；
1-b) then utilize participle technique, from title obtain identification less than candidate entity word list；
Part-of-speech tagging 1-c) is carried out to candidate entity word, weeding out auxiliary word, number, these do not have the candidate vocabulary of practical significance, then
Verify whether these candidate words are entity words on encyclopaedia website, using entity word and the name entity of extraction as kind of a fructification.
3. knowledge mapping dynamic updating method according to claim 1, which is characterized in that according to following in the step 3)
Feature construction training dataset:
Number of weeks existing for 3-1): entity is present in the time on encyclopaedia website, and new entity is more likely to be updated；
3-2) update times: the number that entity is updated in total, the number being updated in the past is more, and future may more continue more
Newly；
3-3) browsing time: the number that an entity is seen is more, illustrates more welcome, is also more likely to be updated；
3-4) number of links: easier to be provided by other when the hyperlink that a physical page includes other resources or entity is more
Source or entity influence；
3-5) internal chaining number: if being linked to other entities, the change of other entities may also will affect oneself；
3-6) content of pages length: an article is longer, and content is abundanter, easier to be updated；
3-7) clip Text length: abstract contains the description most succinct to entity, makes a summary longer, information content is bigger；
3-8) average all renewal frequency: past updates more frequent, then future may also update it is frequent.
4. knowledge mapping dynamic updating method according to claim 1, which is characterized in that according to as follows in the step 5)
Specific steps are updated knowledge mapping:
Firstly, more content of the new seed entity in knowledge mapping, then updates extension entity in knowledge graph according to priority
Content in spectrum: it for the entity being originally not present in knowledge mapping, is inserted into knowledge mapping with highest priority；For that
Entity existing for a little scripts, provides forecast updating frequency first with fallout predictor, then preferentially updates to upper subsynchronous expectation later
The maximum extension entity of frequency is updated, until all extension entities have all been updated completion or extension entity update quantity
The extension entity for reaching setting updates the upper limit of the number value K, and wherein K is determined according to network bandwidth and computing resource.
5. knowledge mapping dynamic updating method according to claim 4, which is characterized in that when the extension entity updates
The priority priority value that is calculate by the following formula determine that priority value is bigger, then priority is higher:
E [u (x)]=P (x) × (tnow-ts(x))
Wherein, E [u (x)] be to the priority value that is updated of extension entity, u (x) be it is upper it is subsynchronous after renewal time, P
It (x) is the renewal frequency predicted by fallout predictor, tnowFor current time, tsIt (x) is the time updated entity x last time, when x is
When novel entities, ts(x)=- ∞.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810627957.7A CN109033160A (en) | 2018-06-15 | 2018-06-15 | A kind of knowledge mapping dynamic updating method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810627957.7A CN109033160A (en) | 2018-06-15 | 2018-06-15 | A kind of knowledge mapping dynamic updating method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109033160A true CN109033160A (en) | 2018-12-18 |
Family
ID=64609573
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810627957.7A Pending CN109033160A (en) | 2018-06-15 | 2018-06-15 | A kind of knowledge mapping dynamic updating method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109033160A (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109800288A (en) * | 2019-01-22 | 2019-05-24 | 杭州师范大学 | A kind of the scientific research analysis of central issue and prediction technique of knowledge based map |
| CN110008353A (en) * | 2019-04-09 | 2019-07-12 | 福建奇点时空数字科技有限公司 | A kind of construction method of dynamic knowledge map |
| CN110162640A (en) * | 2019-04-28 | 2019-08-23 | 北京百度网讯科技有限公司 | Novel entities method for digging, device, computer equipment and storage medium |
| CN110598002A (en) * | 2019-08-14 | 2019-12-20 | 广州视源电子科技股份有限公司 | Knowledge graph library construction method and device, computer storage medium and electronic equipment |
| CN110851612A (en) * | 2019-08-29 | 2020-02-28 | 国家计算机网络与信息安全管理中心 | Encyclopedic knowledge-based mobile application knowledge graph composite completion method and device |
| CN111488467A (en) * | 2020-04-30 | 2020-08-04 | 北京建筑大学 | Construction method and device of geographical knowledge graph, storage medium and computer equipment |
| CN111626821A (en) * | 2020-05-26 | 2020-09-04 | 山东大学 | Product recommendation method and system for realizing customer classification based on integrated feature selection |
| CN111859238A (en) * | 2020-07-27 | 2020-10-30 | 平安科技（深圳）有限公司 | Method and device for predicting data change frequency based on model and computer equipment |
| CN112035682A (en) * | 2020-09-28 | 2020-12-04 | 上海松鼠课堂人工智能科技有限公司 | Character model analysis processing method and system based on knowledge graph |
| CN113065353A (en) * | 2021-03-16 | 2021-07-02 | 北京金堤征信服务有限公司 | Entity identification method and device |
| CN113239208A (en) * | 2021-05-06 | 2021-08-10 | 广东博维创远科技有限公司 | Mark training model based on knowledge graph |
| CN113553436A (en) * | 2020-04-23 | 2021-10-26 | 广东博智林机器人有限公司 | Knowledge graph updating method and device, electronic equipment and storage medium |
| CN113609309A (en) * | 2021-08-16 | 2021-11-05 | 脸萌有限公司 | Knowledge graph construction method and device, storage medium and electronic equipment |
| CN115408534A (en) * | 2022-08-23 | 2022-11-29 | 连连银通电子支付有限公司 | Knowledge graph updating method, device, equipment and storage medium |
| CN116502711A (en) * | 2023-06-27 | 2023-07-28 | 北京智谱华章科技有限公司 | Knowledge graph construction and dynamic expansion method, device, equipment and medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106503035A (en) * | 2016-09-14 | 2017-03-15 | 海信集团有限公司 | A kind of data processing method of knowledge mapping and device |
| CN107346346A (en) * | 2017-08-26 | 2017-11-14 | 海南大学 | Learner competencies modeling and learning process Optimal Management System based on data collection of illustrative plates, Information Atlas and knowledge mapping |
-
2018
- 2018-06-15 CN CN201810627957.7A patent/CN109033160A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106503035A (en) * | 2016-09-14 | 2017-03-15 | 海信集团有限公司 | A kind of data processing method of knowledge mapping and device |
| CN107346346A (en) * | 2017-08-26 | 2017-11-14 | 海南大学 | Learner competencies modeling and learning process Optimal Management System based on data collection of illustrative plates, Information Atlas and knowledge mapping |
Non-Patent Citations (1)
| Title |
|---|
| JIAQING LIANG 等: "How to Keep a Knowledge Base Synchronized with Its Encyclopedia Source", 《PROCEEDINGS OF THE TWENTY-SIXTH INTERNATIONAL JOINT CONFERENCE ON ARTIFICIAL INTELLIGENCE (IJCAI-17)》 * |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109800288A (en) * | 2019-01-22 | 2019-05-24 | 杭州师范大学 | A kind of the scientific research analysis of central issue and prediction technique of knowledge based map |
| CN110008353A (en) * | 2019-04-09 | 2019-07-12 | 福建奇点时空数字科技有限公司 | A kind of construction method of dynamic knowledge map |
| CN110162640A (en) * | 2019-04-28 | 2019-08-23 | 北京百度网讯科技有限公司 | Novel entities method for digging, device, computer equipment and storage medium |
| CN110598002A (en) * | 2019-08-14 | 2019-12-20 | 广州视源电子科技股份有限公司 | Knowledge graph library construction method and device, computer storage medium and electronic equipment |
| CN110851612A (en) * | 2019-08-29 | 2020-02-28 | 国家计算机网络与信息安全管理中心 | Encyclopedic knowledge-based mobile application knowledge graph composite completion method and device |
| CN110851612B (en) * | 2019-08-29 | 2023-08-18 | 国家计算机网络与信息安全管理中心 | Mobile application knowledge graph composite completion method and device based on encyclopedia knowledge |
| CN113553436A (en) * | 2020-04-23 | 2021-10-26 | 广东博智林机器人有限公司 | Knowledge graph updating method and device, electronic equipment and storage medium |
| CN111488467A (en) * | 2020-04-30 | 2020-08-04 | 北京建筑大学 | Construction method and device of geographical knowledge graph, storage medium and computer equipment |
| CN111488467B (en) * | 2020-04-30 | 2022-04-05 | 北京建筑大学 | Construction method and device of geographical knowledge graph, storage medium and computer equipment |
| CN111626821A (en) * | 2020-05-26 | 2020-09-04 | 山东大学 | Product recommendation method and system for realizing customer classification based on integrated feature selection |
| CN111626821B (en) * | 2020-05-26 | 2024-03-12 | 山东大学 | Product recommendation method and system for realizing customer classification based on integrated feature selection |
| CN111859238A (en) * | 2020-07-27 | 2020-10-30 | 平安科技（深圳）有限公司 | Method and device for predicting data change frequency based on model and computer equipment |
| CN111859238B (en) * | 2020-07-27 | 2024-07-16 | 平安科技（深圳）有限公司 | Model-based method, device and computer equipment for predicting data change frequency |
| CN112035682A (en) * | 2020-09-28 | 2020-12-04 | 上海松鼠课堂人工智能科技有限公司 | Character model analysis processing method and system based on knowledge graph |
| CN113065353A (en) * | 2021-03-16 | 2021-07-02 | 北京金堤征信服务有限公司 | Entity identification method and device |
| CN113065353B (en) * | 2021-03-16 | 2024-04-02 | 北京金堤征信服务有限公司 | Entity identification method and device |
| CN113239208A (en) * | 2021-05-06 | 2021-08-10 | 广东博维创远科技有限公司 | Mark training model based on knowledge graph |
| CN113609309A (en) * | 2021-08-16 | 2021-11-05 | 脸萌有限公司 | Knowledge graph construction method and device, storage medium and electronic equipment |
| CN113609309B (en) * | 2021-08-16 | 2024-02-06 | 脸萌有限公司 | Knowledge graph construction method and device, storage medium and electronic equipment |
| WO2023022655A3 (en) * | 2021-08-16 | 2023-04-13 | 脸萌有限公司 | Knowledge map construction method and apparatus, storage medium, and electronic device |
| CN115408534B (en) * | 2022-08-23 | 2023-12-12 | 连连银通电子支付有限公司 | Knowledge graph updating method, device, equipment and storage medium |
| CN115408534A (en) * | 2022-08-23 | 2022-11-29 | 连连银通电子支付有限公司 | Knowledge graph updating method, device, equipment and storage medium |
| CN116502711B (en) * | 2023-06-27 | 2023-09-26 | 北京智谱华章科技有限公司 | Knowledge graph construction and dynamic expansion method, device, equipment and medium |
| CN116502711A (en) * | 2023-06-27 | 2023-07-28 | 北京智谱华章科技有限公司 | Knowledge graph construction and dynamic expansion method, device, equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109033160A (en) | A kind of knowledge mapping dynamic updating method | |
| CN103176985B (en) | The most efficient a kind of internet information crawling method | |
| CN106777274B (en) | A kind of Chinese tour field knowledge mapping construction method and system | |
| US7424484B2 (en) | Path-based ranking of unvisited web pages | |
| CN102902806B (en) | A kind of method and system utilizing search engine to carry out query expansion | |
| CN102567464B (en) | Based on the knowledge resource method for organizing of expansion thematic map | |
| CN110704624B (en) | Geographic information service metadata text multi-level multi-label classification method | |
| CN107169001A (en) | A kind of textual classification model optimization method based on mass-rent feedback and Active Learning | |
| CN104866557B (en) | A kind of personalized instant learning theoretical based on constructive learning supports System and method for | |
| CN107577759A (en) | User comment auto recommending method | |
| CN103955842B (en) | A kind of online advertisement commending system and method towards mass media data | |
| CN106156372B (en) | A kind of classification method and device of internet site | |
| CN101350011A (en) | Method for detecting search engine cheat based on small sample set | |
| CN105718579A (en) | Information push method based on internet-surfing log mining and user activity recognition | |
| CN103744956B (en) | A kind of diversified expanding method of key word | |
| CN106204156A (en) | A kind of advertisement placement method for network forum and device | |
| CN107885793A (en) | A kind of hot microblog topic analyzing and predicting method and system | |
| CN102708096A (en) | Network intelligence public sentiment monitoring system based on semantics and work method thereof | |
| CN106066873A (en) | A kind of travel information based on body recommends method | |
| CN106227788A (en) | Database query method based on Lucene | |
| CN103049542A (en) | Domain-oriented network information search method | |
| CN104182412A (en) | Webpage crawling method and webpage crawling system | |
| CN108710672B (en) | Theme crawler method based on incremental Bayesian algorithm | |
| CN107833142A (en) | Academic social networks scientific research cooperative person recommends method | |
| CN104239496A (en) | Collaborative filtering method based on integration of fuzzy weight similarity measurement and clustering |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181218 |