CN109460479A - A kind of prediction technique based on reason map, device and system - Google Patents
A kind of prediction technique based on reason map, device and system Download PDFInfo
- Publication number
- CN109460479A CN109460479A CN201811375960.0A CN201811375960A CN109460479A CN 109460479 A CN109460479 A CN 109460479A CN 201811375960 A CN201811375960 A CN 201811375960A CN 109460479 A CN109460479 A CN 109460479A
- Authority
- CN
- China
- Prior art keywords
- resume
- event
- network
- data
- unit
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/10—Office automation; Time management
- G06Q10/105—Human resources
- G06Q10/1053—Employment or hiring
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Machine Translation (AREA)
Abstract
The present invention provides a kind of prediction techniques based on reason map, device and system, and wherein method includes: acquisition resume inventory；Data parsing is carried out to the resume data in database；Event Distillation is carried out to the data after parsing, forms resume event network；It according to the information to be predicted received, is predicted in the resume event network, and the information of prediction is shown.It can carry out carrying out talent's occupational planning using technical solution of the present invention, resume recommendation is carried out to robot mechanism, the offers references such as intention work recommendation are carried out to the talent, improve user's operation experience and interview success rate, while objective, comprehensive information is provided, reduce time and manual read's information the time it takes of screening and analysis.
Description
Technical field
The present invention relates to technical field of data processing, in particular to a kind of prediction technique based on reason map, dress
It sets and system.
Background technique
For current knowledge mapping using attribute, relationship as main study subject, attribute and relationship can be attributed to Subject, Predicate and Object
(SPO) triple relationship.Resume knowledge mapping is no exception.I.e. resume knowledge mapping is to grind with the relationship between concept and concept
Study carefully object.
Event is one of key concept of human society, and the social activities of people is often event driven, and resume is not yet
Exception.Between event in time sequencing along hold generation, the causality between front and back, when selecting next event it is eventful
The probability feature of part, formed when multiple affair forms an event chain a whole event and whole event with it is whole
Evolution and mode between body event etc. are a kind of very valuable knowledge.
Reason map (Event evolutionary graph) is a reason and logic knowledge base, between description event
Evolution and mode.Reason map defines a variety of interevent relations, such as: suitable to hold relationship, causality etc..Substantially reason
Map is a reason and logic knowledge base, describes Evolution and mode between event, can apply in life very
Various aspects, such as event prediction, commonsense reasoning, consumption are intended to excavate, and dialogue generates etc..
However, existing most of knowledge mappings all do not extract event, especially in domain-specific knowledge map
On, such as resume knowledge mapping.Existing part resume knowledge mapping is formed in form by keyword, the method for template matching
Resume event map to a certain degree.
It is above-mentioned in the prior art, resume knowledge mapping by keyword, template matching method carry out resume event map
Building, there are apparent defects.Its building method be usually scholar's manual coding, manual construction mode complete.This method
Compare mechanization, need constructor that there is the knowledge of specific area, need to construct a large amount of keyword, a large amount of artificial template,
Once word is different, expression way is different, existing keyword, template will be no longer applicable in, and long term maintenance keyword, mould
The cost of labor of plate is very high.
Summary of the invention
It is an object of the invention to by carrying out data extraction to resume inventory, resume data time letter is extracted in automation
It ceases, and learns, trains resume event network --- the resume occurrence diagram of an event Evolution using deep neural network
Spectrum carries out the operation such as event prediction further according to resume event network.It can carry out carrying out talent's duty using technical solution of the present invention
Industry planning, carries out resume recommendation to robot mechanism, carries out the offers references such as intention work recommendation to the talent, improves user's operation
Experience and interview success rate, decrease the time of screening and analysis.
To achieve the goals above, technical solution of the present invention provides a kind of prediction technique based on reason map, packet
It includes: obtaining resume inventory；Data parsing is carried out to the resume data in database；Event is carried out to the data after parsing to mention
It takes, forms resume event network；It according to the information to be predicted received, is predicted in the resume event network, and will
The information of prediction is shown.
Specifically, by obtaining resume inventory, the data of resume could be preferably obtained, more accurately letter could be generated
Go through event network.The mode that resume obtains has very much, first can be in benefits such as some recruitment websites, GitHub, forum, University Websites
Technology, which is crawled, with network carries out data acquisition.Secondly it can cooperate with headhunter, it is standby to obtain document, JSON, database
The resume of part format.Be finally directly with recruitment website such as intelligence connection recruitment, future is carefree etc., and websites are cooperated, backstage directly
Obtain resume data.The resume data of these platforms are typically stored in relevant database, belong to semi-structured data, can be with
These initial data are replicated a for further part use.
Specifically, data parsing is carried out for the resume data in database, for extracting the information in resume data.Its
In, general resume has following several module informations: essential information, education background, work experience, project experiences, self-assessment, language
Speech ability, training information, reward situation, technical capability etc..Wherein the known module for having event that can extract be work experience,
Three project experiences, self-assessment modules.It is therefore desirable to which the information of these three modules is split.If it is on recruitment website
The resume of acquisition, these three module segmentations are fairly simple, have divided module on the backstage of these websites, can directly extract
The information of these modules pieces together several little modules.If it is the resume of document form, need according to template matching
It is split such as regular expression or by online some resume resolvers.
In the present solution, the data after described pair of parsing carry out Event Distillation, resume event network is formed, comprising: to parsing
Data afterwards carry out sentence segmentation, form small sentence unit；Interdependent syntactic analysis is carried out for the small sentence unit, obtains thing
Part information forms resume event network.
Specifically, different neighborhood event sentence dividing methods are different, can according to comma, pause mark, branch, fullstop, colon,
Every kind of symbol of the punctuation marks such as question mark, newline is all used as separator to carry out small sentence division.Can also only with fullstop, emit
Number, newline carry out big sentence division.Resume event sentence is divided with comma, pause mark, branch, fullstop, colon, question mark, line feed
Every kind of symbol of the punctuation marks such as symbol is all used as separator to carry out small sentence division.Above-mentioned partitioning scheme can not limit this hair
Bright protection scope, those skilled in the art should be understood that any method that can be realized sentence segmentation should all fall into the present invention
In protection scope.
In the present solution, described carry out interdependent syntactic analysis for the small sentence unit, event information is obtained, comprising: base
Interdependent syntactic analysis is carried out in maximum entropy model and maximum spanning tree model or carries out interdependent syntax based on conditional random field models
Analysis, obtains event information.
In the present solution, the formation resume event network, comprising: carried out at participle by participle technique to event information
Reason, then switchs to vector for participle according to trained word2vec model；The vector of all participles is merged into a vector；
Based on gate figure neural network GGNN algorithm, resume event network is generated.
In the present solution, the vector by all participles merges into a vector, comprising:
Based on one or more of average algorithm, linear transformation algorithm, stitching algorithm, the vector of all participles is closed
It and is a vector.
Second aspect of the present invention also provides a kind of prediction meanss based on reason map, comprising:
Acquiring unit, for obtaining resume inventory；Resolution unit, for carrying out data to the resume data in database
Parsing；Construction unit forms resume event network for carrying out Event Distillation to the data after parsing；Predicting unit, according to connecing
The information to be predicted received is predicted in the resume event network, and the information of prediction is shown.
Specifically, by obtaining resume inventory, the data of resume could be preferably obtained, more accurately letter could be generated
Go through event network.The mode that resume obtains has very much, first can be in benefits such as some recruitment websites, GitHub, forum, University Websites
Technology, which is crawled, with network carries out data acquisition.Secondly it can cooperate with headhunter, it is standby to obtain document, JSON, database
The resume of part format.Be finally directly with recruitment website such as intelligence connection recruitment, future is carefree etc., and websites are cooperated, backstage directly
Obtain resume data.The resume data of these platforms are typically stored in relevant database, belong to semi-structured data, can be with
These initial data are replicated a for further part use.
Specifically, data parsing is carried out for the resume data in database, for extracting the information in resume data.Its
In, general resume has following several module informations: essential information, education background, work experience, project experiences, self-assessment, language
Speech ability, training information, reward situation, technical capability etc..Wherein the known module for having event that can extract be work experience,
Three project experiences, self-assessment modules.It is therefore desirable to which the information of these three modules is split.If it is recruitment website,
GitHub, forum, University Websites, the resume obtained on recruitment website, these three module segmentations are fairly simple, behind these websites
Platform has divided module, can directly extract the information of these modules or piece together several little modules.If it is text
The resume of shelves form, needs to be divided according to template matching such as regular expression or by online some resume resolvers
It cuts.
In the present solution, the construction unit, comprising: cutting unit, for carrying out sentence segmentation, shape to the data after parsing
At small sentence unit；Analytical unit obtains event information, shape for carrying out interdependent syntactic analysis for the small sentence unit
At resume event network.
Specifically, different neighborhood event sentence dividing methods are different, can according to comma, pause mark, branch, fullstop, colon,
Every kind of symbol of the punctuation marks such as question mark, newline is all used as separator to carry out small sentence division.Can also only with fullstop, emit
Number, newline carry out big sentence division.Resume event sentence is divided with comma, pause mark, branch, fullstop, colon, question mark, line feed
Every kind of symbol of the punctuation marks such as symbol is all used as separator to carry out small sentence division.Above-mentioned partitioning scheme can not limit this hair
Bright protection scope, those skilled in the art should be understood that any method that can be realized sentence segmentation should all fall into the present invention
In protection scope.
In the present solution, the analytical unit, comprising: carry out interdependent syntax based on maximum entropy model and maximum spanning tree model
Analysis carries out interdependent syntactic analysis based on conditional random field models, obtains event information.
In the present solution, the construction unit, comprising: conversion unit, for being divided by participle technique event information
Word processing, then switchs to vector for participle according to trained word2vec model；Combining unit, for by all participles to
Amount merges into a vector；Computing unit, for generating resume event network based on gate figure neural network GGNN algorithm.
In the present solution, the vector by all participles merges into a vector, comprising:
Based on one or more of average algorithm, linear transformation algorithm, stitching algorithm, the vector of all participles is closed
It and is a vector.
Third aspect present invention also provides a kind of talents information recommender system, the system comprises: memory, processor and
It is stored in the program for the prediction technique based on reason map that can be run on the memory and on the processor, the base
Such as the above-mentioned prediction side based on reason map is realized when the program of the prediction technique of reason map is executed by the processor
The step of method.
For the present invention by carrying out data extraction to resume inventory, resume data time information is extracted in automation, and is utilized
Deep neural network study, resume event network --- the resume event map for training an event Evolution, further according to
Resume event network carries out the operation such as event prediction, realizes the forecast function to event.It can using technical solution of the present invention
It carries out carrying out talent's occupational planning, carries out resume recommendation to robot mechanism, carry out the offers references such as intention work recommendation to the talent,
User's operation experience and interview success rate are improved, while objective, comprehensive information is provided, reduces screening and analysis
Time and manual read's information the time it takes.
Additional aspect and advantage of the invention will provide in following description section, will partially become from the following description
Obviously, or practice through the invention is recognized.
Detailed description of the invention
Above-mentioned and/or additional aspect of the invention and advantage will become from the description of the embodiment in conjunction with the following figures
Obviously and it is readily appreciated that, in which:
Fig. 1 shows a kind of flow chart of the prediction technique based on reason map of the present invention；
Fig. 2 shows interdependent syntactic analysis schematic diagrames of the invention；
Fig. 3 shows the relational graph of the interdependent syntactic analysis of the present invention；
Fig. 4 shows to form resume event network method flow chart；
Fig. 5 shows a kind of block diagram of the prediction meanss based on reason map of the present invention.
Specific embodiment
To better understand the objects, features and advantages of the present invention, with reference to the accompanying drawing and specific real
Applying mode, the present invention is further described in detail.It should be noted that in the absence of conflict, the implementation of the application
Feature in example and embodiment can be combined with each other.
In the following description, numerous specific details are set forth in order to facilitate a full understanding of the present invention, still, the present invention may be used also
To be implemented using other than the one described here other modes, therefore, protection scope of the present invention is not by described below
Specific embodiment limitation.
Fig. 1 shows a kind of flow chart of the prediction technique based on reason map of the present invention.
As shown in Figure 1, the prediction technique based on reason map of the embodiment the following steps are included:
S102 obtains resume inventory；
S104 carries out data parsing to the resume data in database；
S106 carries out Event Distillation to the data after parsing, forms resume event network；
S108 predicted in the resume event network according to the information to be predicted received, and by the letter of prediction
Breath is shown.
In the present embodiment, by obtaining resume inventory, the data of resume could be preferably obtained, could be generated more acurrate
Resume event network.The mode that resume obtains has very much, first can be in some recruitment websites, GitHub, forum, University Websites
Data acquisition is carried out Deng technology is crawled using network.Secondly it can cooperate with headhunter, obtain document, JSON, data
The resume of library backup format.Be finally directly with recruitment website such as intelligence connection recruitment, future is carefree etc., and websites are cooperated, on backstage
Directly acquire resume data.The resume data of these platforms are typically stored in relevant database, belong to semi-structured data,
These initial data can be replicated a for further part use.
In the present embodiment, data parsing is carried out for the resume data in database, for extracting the letter in resume data
Breath.Wherein, general resume has following several module informations: essential information, work experience, project experiences, self is commented education background
Valence, language competence, training information, reward situation, technical capability etc..Wherein the known module for having event that can extract is work
Three experience, project experiences, self-assessment modules.It is therefore desirable to which the information of these three modules is split.If it is recruitment
Net, GitHub, forum, University Websites, the resume obtained on recruitment website, these three module segmentations are fairly simple, in these websites
Backstage divided module, can directly extract the information of these modules or piece together several little modules.If
It is the resume of document form, needs to carry out according to template matching such as regular expression or by online some resume resolvers
Segmentation.
It should be noted that also need to polymerize different titles in the data resolving of resume inventory,
To obtain unified name information.
Wherein, also need to realize the polymerization of school's title in resume inventory, school's title that different people writes may
It is different, for example Peking University, someone write Peking University, Beijing University or Peking University.What these three were said the name of sth.
All it is the same school, i.e. Peking University, needs to be polymerized to the same entity.By crawler and search technique from education network or its
The full school's title of some domestic and international comparisons is downloaded in its website, forms standard school namebase, then such as using similarity algorithm
TF-IDF (the reverse document of word frequency -) scheduling algorithm compares the similarity of school's title and standard school title in resume inventory, choosing
The highest standard school title of similarity is selected as corresponding school's title in resume data, is learned to reach polymerization
The purpose of school entity.
Also needing to realize the polymerization of Business Name in resume inventory, the Business Name that different people writes is also different,
For example Jingdone district store, somebody write a Chinese character in simplified form Jingdone district store, Jingdone district, 360buy, Beijing 360buy Co., Ltd., Beijing Jingdone district
Store, the Jingdone district 360buy store etc..By crawler and search technique, more complete company is carried from some business directories are off the net
Title forms Standard Co., Ltd's name database, then utilizes similarity algorithm such as TF-IDF (the reverse document of word frequency -) scheduling algorithm ratio
Compared with the similarity of Business Name in resume inventory and Standard Co., Ltd's title, select similarity highest one as resume data
In corresponding Business Name, thus achieve the purpose that polymerize corporate entity.For some company names for there are several titles, such as happily gather
Epoch, YY, more objects for appreciation, what is referred to is all same company, and needing to construct equivalence relation makes different titles be directed toward the same entity.
In the present solution, the data after described pair of parsing carry out Event Distillation, resume event network is formed, comprising: to parsing
Data afterwards carry out sentence segmentation, form small sentence unit；Interdependent syntactic analysis is carried out for the small sentence unit, obtains thing
Part information forms resume event network.
It should be noted that different neighborhood event sentence dividing methods are different, it can be according to comma, pause mark, branch, sentence
Number, every kind of symbol of the punctuation marks such as colon, question mark, newline be all used as separator to carry out small sentence division.Can also only with
Fullstop, colon, newline carry out big sentence division.Resume event sentence is divided with comma, pause mark, branch, fullstop, colon, is asked
Number, every kind of symbol of the punctuation marks such as newline be all used as separator to carry out small sentence division.Above-mentioned partitioning scheme can not
It limits the scope of the invention, those skilled in the art should be understood that any method that can be realized sentence segmentation should all be fallen
Enter in the scope of the present invention.
Extracting the event stage, as shown in Fig. 2, Event Distillation mainly extracts the Subject, Predicate and Object (SPO) of event, need to use according to
Deposit syntactic analysis technology.Relationship type, TAG, description when Fig. 3 shows interdependent syntactic analysis.Interdependent syntactic analysis is mainly
The structure of parsing sentence.
According to embodiments of the present invention, described to carry out interdependent syntactic analysis for the small sentence unit, event information is obtained,
Include: based on maximum entropy model and maximum spanning tree model carry out interdependent syntactic analysis or based on conditional random field models carry out according to
Syntactic analysis is deposited, event information is obtained.
It should be noted that carrying out interdependent syntactic analysis based on maximum entropy model and maximum spanning tree model, event is obtained
Information, specifically:
Maximum entropy model: maximum entropy model (Maximum Entropy Model, hereinafter referred to as MaxEnt), MaxEnt is
A criterion in probabilistic model study, thought are as follows: in learning probability model, all possible maximum model of model medium entropy
It is best model；If probabilistic model needs to meet some constraints, principle of maximum entropy is exactly in the condition for meeting known constraints
Selective entropy maximum model in set.Principle of maximum entropy points out that when predicting the probability distribution of a chance event, prediction is answered
When satisfaction all known constraints, and any subjective hypothesis is not done to unknown situation.In this case, probability distribution is most
Uniformly, the least risk of prediction, therefore the entropy of obtained probability distribution is maximum.
Maximum spanning tree model: maximum spanning tree model (maximum spanning trees, MST) defines whole syntax
The marking of tree is the weighted sum of each side marking in tree:
S indicates marking value, and y is a dependency tree of sentence x, (i, j) be two words in y constitute a line (according to
Deposit relationship), f is the higher-dimension binary feature functional vector that value is 1 or 0, indicates to whether there is interdependent pass between node xi and xj
System.W is the weight vector of feature f (i, j), and w is obtained after feature has been determined by sample training.
It should be noted that carrying out interdependent syntactic analysis based on conditional random field models, event information is obtained specifically:
CRF (condition random field) is common model in sequence labelling scene, can be utilized than HMM (Hidden Markov Model)
More features, the problem of than MEMM (maximum entropy Markov model) more resistant against marking bias.This is one based on CRF
The characteristic function of Chinese dependency parsing device, internal CRF model is stored up using even numbers group Trie tree (DoubleArrayTrie)
It deposits, decoding uses the Viterbi backward algorithm of specialization.
After having carried out interdependent syntactic analysis, there is dependency grammar tree, so that it may according to the structure extraction sentence of tree
Subject, Predicate and Object, to form event.One section of word or the event of several sentences composition have along hold, the event chain or thing of the relationships such as cause and effect
Part network.
Such as: " I is responsible for intelligent contract exploitation design work ", it is after Event Distillation program the result is that: ' O':
[' exploitation ', ' design ', ' work '], ' S':[' I '], ' P':[' be responsible for '].Wherein subject is " I ", and predicate is " responsible ",
Object is " design ", " exploitation ", " work ".
Such as Fig. 4, according to embodiments of the present invention, the formation resume event network, comprising: S402 is by participle technique to thing
Part information carries out word segmentation processing, and participle is then switched to vector according to trained word2vec model；S404 is by all participles
Vector merge into a vector；S406 is based on gate figure neural network GGNN algorithm, generates resume event network.
It should be noted that have a large amount of event and its front and back along hold, causality, so that it may construct Context event net
Network.Just similar Google is the same in building word2vec model for Context event network, in word2vec model, model handle
Word is mapped in a higher dimensional space, and a point in space represents a word, and the distance of two different points can define him
Semantic dependency or similitude." emperor " is relevant, the distance of " man " and " woman " also phase at a distance from " queen "
It closes.So (" emperor "-" man ") ≈ (" queen "-" woman ").In simple terms, word can also be added and subtracted as number, from
And distance is provided with semantic relation, here it is so-called semantic nets.Similarly, semantic after the training of magnanimity course of event semantic net
Similar event indicates with similar vector, event mutually add and subtract can be constituted with reasoning it is suitable hold, causal next event.
It first passes through participle tool to segment the Subject, Predicate and Object (SPO) of event, then according to trained word2vec mould
Word is switched to vector by type.The vector of all words of Subject, Predicate and Object (SPO) is finally merged into a vector to complete to a thing
The vector of part indicates.
In the present solution, the vector by all participles merges into a vector, comprising:
Based on one or more of average algorithm, linear transformation algorithm, stitching algorithm, the vector of all participles is closed
It and is a vector.
Mean value method:
I.e. all term vectors are cumulative to average.
Linear transformation:
Similar single layer perceptron method.
ve=tanh (Wp·vp+W0·va0+W1·va1+W2·va2+b)
Wherein W, b are model parameters.
Splicing method:
I.e. institute's directed quantity is spliced into the higher vector of dimension.
Preferably, the vector of all participles is merged into a vector using linear conversion method.It should be noted that this
The algorithm above can also be combined according to actual needs and merge into a vector by field technical staff.
Directed cyclic graph is event relation network, is similar to word2vec network.Here event network is not chain knot
Structure, but the network structure that directed acyclic graph is constituted.Network is by GGNN (gate figure neural network, gated graph neural
Network it) realizes.Gated are realized using GRU (gated recurrent units).There are two matrixes by GRU, and one is h
Matrix indicates the matrix that event context and successor are constituted:
h(0)={ ve1, ve2..., ven, vec1, vec2..., veck}
Wherein ve1, ve2..., venIt is context events, vec1, vec2..., veckNext candidate events list has k
Event is available.
The other is adjacency matrix A, using individual event V as node, the side E between node and node be used as front and back because
Fruit, along hold, probabilistic relation.
Adjacency matrix has recorded the correlation between event V.The process of GRU network is as follows:
a(t)=ATh(t-1)+b
z(t)=σ (Wza(t)+U2h(t-1)
r(t)=σ (Wra(t)+Urh(t-1))
After a large amount of corpus carries out network training, network parameter tends to stablize, so as to form resume thing
Part " knowledge " network.
After forming resume event network, event prediction can be carried out.According to the information to be predicted received, in institute
It states and is predicted in resume event network, and the information of prediction is shown.
It should be noted that volume can receive incoming event information after training above-mentioned resume event " knowledge " network,
It is predicted.It specifically includes: input ve1, ve2..., venIt is context events, obtains the optional sequence v of next eventec1,
vec2..., veck.The relevance function s of front and back event can be defined:
Sij=g (Vi (t), Vcj (t))
Relevance function can be replaced with similarity function.Similarity can be defined so that there are as below methods:
Manhattan similarity
Calculate the distance of two vectors:
manhattan(Vi (t), Vcj (t))=∑ | Vi (t)-Vcj (t)|
Cosine display degree
Inner product similarity
Euclidean distance
This patent uses Euclidean distance.After the similarity relationship sij for having context events and candidate events, then
It averages to single candidate events all values, it may be assumed that
Maximum value sj is asked to obtain predicted events all candidate events:
C=maxjsj。
Technical solution of the present invention can automatically extract event chain after inputting resume original text.Resume library based on magnanimity mentions
After taking event, learn and train the resume event network of magnanimity.Have resume event network, can based on existing event into
The prediction of the next event of row, to carry out occupational planning for the talent, to carrying out resume recommendation with robot mechanism, anticipate to the talent
Recommend to work etc..
As shown in figure 5, a kind of prediction meanss based on reason map of the embodiment, comprising:
Acquiring unit 502, for obtaining resume inventory；Resolution unit 504, for the resume data in database into
The parsing of row data；Construction unit 506 forms resume event network for carrying out Event Distillation to the data after parsing；Prediction is single
Member 508, according to the information to be predicted received, is predicted in the resume event network, and the information of prediction is carried out
It shows.
In the present embodiment, by obtaining resume inventory, the data of resume could be preferably obtained, could be generated more acurrate
Resume event network.The mode that resume obtains has very much, first can be in some recruitment websites, GitHub, forum, University Websites
Data acquisition is carried out Deng technology is crawled using network.Secondly it can cooperate with headhunter, obtain document, JSON, data
The resume of library backup format.Be finally directly with recruitment website such as intelligence connection recruitment, future is carefree etc., and websites are cooperated, on backstage
Directly acquire resume data.The resume data of these platforms are typically stored in relevant database, belong to semi-structured data,
These initial data can be replicated a for further part use.
In the present embodiment, data parsing is carried out for the resume data in database, for extracting the letter in resume data
Breath.Wherein, general resume has following several module informations: essential information, work experience, project experiences, self is commented education background
Valence, language competence, training information, reward situation, technical capability etc..Wherein the known module for having event that can extract is work
Three experience, project experiences, self-assessment modules.It is therefore desirable to which the information of these three modules is split.If it is recruitment
Net, GitHub, forum, University Websites, the resume obtained on recruitment website, these three module segmentations are fairly simple, in these websites
Backstage divided module, can directly extract the information of these modules or piece together several little modules.If
It is the resume of document form, needs to carry out according to template matching such as regular expression or by online some resume resolvers
Segmentation.
It should be noted that also need to polymerize different titles in the data resolving of resume inventory,
To obtain unified name information.
Wherein, also need to realize the polymerization of school's title in resume inventory, school's title that different people writes may
It is different, for example Peking University, someone write Peking University, Beijing University or Peking University.What these three were said the name of sth.
All it is the same school, i.e. Peking University, needs to be polymerized to the same entity.By crawler and search technique from education network or its
The full school's title of some domestic and international comparisons is downloaded in its website, forms standard school namebase, then such as using similarity algorithm
TF-IDF (the reverse document of word frequency -) scheduling algorithm compares the similarity of school's title and standard school title in resume inventory, choosing
The highest standard school title of similarity is selected as corresponding school's title in resume data, is learned to reach polymerization
The purpose of school entity.
Also needing to realize the polymerization of Business Name in resume inventory, the Business Name that different people writes is also different,
For example Jingdone district store, somebody write a Chinese character in simplified form Jingdone district store, Jingdone district, 360buy, Beijing 360buy Co., Ltd., Beijing Jingdone district
Store, the Jingdone district 360buy store etc..By crawler and search technique, more complete company is carried from some business directories are off the net
Title forms Standard Co., Ltd's name database, then utilizes similarity algorithm such as TF-IDF (the reverse document of word frequency -) scheduling algorithm ratio
Compared with the similarity of Business Name in resume inventory and Standard Co., Ltd's title, select similarity highest one as resume data
In corresponding Business Name, thus achieve the purpose that polymerize corporate entity.For some company names for there are several titles, such as happily gather
Epoch, YY, more objects for appreciation, what is referred to is all same company, and needing to construct equivalence relation makes different titles be directed toward the same entity.
In the present solution, the construction unit 506, comprising: cutting unit 5062, for carrying out sentence to the data after parsing
Segmentation, forms small sentence unit；Analytical unit 5064 is obtained for carrying out interdependent syntactic analysis for the small sentence unit
Event information forms resume event network.
It should be noted that different neighborhood event sentence dividing methods are different, it can be according to comma, pause mark, branch, sentence
Number, every kind of symbol of the punctuation marks such as colon, question mark, newline be all used as separator to carry out small sentence division.Can also only with
Fullstop, colon, newline carry out big sentence division.Resume event sentence is divided with comma, pause mark, branch, fullstop, colon, is asked
Number, every kind of symbol of the punctuation marks such as newline be all used as separator to carry out small sentence division.Above-mentioned partitioning scheme can not
It limits the scope of the invention, those skilled in the art should be understood that any method that can be realized sentence segmentation should all be fallen
Enter in the scope of the present invention.
Extracting the event stage, as shown in Fig. 2, Event Distillation mainly extracts the Subject, Predicate and Object (SPO) of event, need to use according to
Deposit syntactic analysis technology.Relationship type, TAG, description when Fig. 3 shows interdependent syntactic analysis.Interdependent syntactic analysis is mainly
The structure of parsing sentence.
In the present solution, the analytical unit 5064, comprising: carried out based on maximum entropy model and maximum spanning tree model interdependent
Syntactic analysis carries out interdependent syntactic analysis based on conditional random field models, obtains event information.
It should be noted that carrying out interdependent syntactic analysis based on maximum entropy model and maximum spanning tree model, event is obtained
Information, specifically:
Maximum entropy model: maximum entropy model (Maximum Entropy Model, hereinafter referred to as MaxEnt), MaxEnt is
A criterion in probabilistic model study, thought are as follows: in learning probability model, all possible maximum model of model medium entropy
It is best model；If probabilistic model needs to meet some constraints, principle of maximum entropy is exactly in the condition for meeting known constraints
Selective entropy maximum model in set.Principle of maximum entropy points out that when predicting the probability distribution of a chance event, prediction is answered
When satisfaction all known constraints, and any subjective hypothesis is not done to unknown situation.In this case, probability distribution is most
Uniformly, the least risk of prediction, therefore the entropy of obtained probability distribution is maximum.
Maximum spanning tree model: maximum spanning tree model (maximum spanning trees, MST) defines whole syntax
The marking of tree is the weighted sum of each side marking in tree:
S indicates marking value, and y is a dependency tree of sentence x, (i, j) be two words in y constitute a line (according to
Deposit relationship), f is the higher-dimension binary feature functional vector that value is 1 or 0, indicates to whether there is interdependent pass between node xi and xj
System.W is the weight vector of feature f (i, j), and w is obtained after feature has been determined by sample training.
It should be noted that carrying out interdependent syntactic analysis based on conditional random field models, event information is obtained specifically:
CRF (condition random field) is common model in sequence labelling scene, can be utilized than HMM (Hidden Markov Model)
More features, the problem of than MEMM (maximum entropy Markov model) more resistant against marking bias.This is one based on CRF
The characteristic function of Chinese dependency parsing device, internal CRF model is stored up using even numbers group Trie tree (DoubleArrayTrie)
It deposits, decoding uses the Viterbi backward algorithm of specialization.
After having carried out interdependent syntactic analysis, there is dependency grammar tree, so that it may according to the structure extraction sentence of tree
Subject, Predicate and Object, to form event.One section of word or the event of several sentences composition have along hold, the event chain or thing of the relationships such as cause and effect
Part network.
Such as: " I is responsible for intelligent contract exploitation design work ", it is after Event Distillation program the result is that: ' O':
[' exploitation ', ' design ', ' work '], ' S':[' I '], ' P':[' be responsible for '].Wherein subject is " I ", and predicate is " responsible ",
Object is " design ", " exploitation ", " work ".
In the present solution, the construction unit 506, comprising: conversion unit 5066, for passing through participle technique to event information
Word segmentation processing is carried out, participle is then switched to by vector according to trained word2vec model；Combining unit 5068 is used for institute
There is the vector of participle to merge into a vector；Computing unit 5010, for generating letter based on gate figure neural network GGNN algorithm
Go through event network.
It should be noted that have a large amount of event and its front and back along hold, causality, so that it may construct Context event net
Network.Context event network is just as Google is constructing word2vec model, and in word2vec model, model is word
It is mapped in a higher dimensional space, a point in space represents a word, and the distance of two different points can define them
Semantic dependency or similitude." emperor " be at a distance from " queen " it is relevant, " man " and the distance of " woman " are also related
's.So (" emperor "-" man ") ≈ (" queen "-" woman ").In simple terms, word can also be added and subtracted as number, thus
Distance is provided with semantic relation, and here it is so-called semantic nets.Similarly, after the training of magnanimity course of event semantic net, semantic category
As event indicated with similar vector, event mutually add and subtract can be constituted with reasoning it is suitable hold, causal next event.
It first passes through participle tool to segment the Subject, Predicate and Object (SPO) of event, then according to trained word2vec mould
Word is switched to vector by type.The vector of all words of Subject, Predicate and Object (SPO) is finally merged into a vector to complete to a thing
The vector of part indicates.
In the present solution, the vector by all participles merges into a vector, comprising:
Based on one or more of average algorithm, linear transformation algorithm, stitching algorithm, the vector of all participles is closed
It and is a vector.
Mean value method:
I.e. all term vectors are cumulative to average.
Linear transformation:
Similar single layer perceptron method.
ve=tanh (Wp·vp+W0·va0+W1·va1+W2·va2+b)
Wherein W, b are model parameters.
Splicing method:
I.e. institute's directed quantity is spliced into the higher vector of dimension.
Preferably, the vector of all participles is merged into a vector using linear conversion method.It should be noted that this
The algorithm above can also be combined according to actual needs and merge into a vector by field technical staff.
Directed cyclic graph is event relation network, is similar to word2vec network.Here event network is not chain knot
Structure, but the network structure that directed acyclic graph is constituted.Network is by GGNN (gate figure neural network, gated graph neural
Network it) realizes.Gated are realized using GRU (gated recurrent units).There are two matrixes by GRU, and one is h
Matrix indicates the matrix that event context and successor are constituted:
h(0)={ ve1, ve2..., ven, vec1, vec2..., veck}
Wherein ve1, ve2..., venIt is context events, vec1, vec2..., veckNext candidate events list has k
Event is available.
The other is adjacency matrix A, using individual event V as node, the side E between node and node be used as front and back because
Fruit, along hold, probabilistic relation.
Adjacency matrix has recorded the correlation between event V.The process of GRU network is as follows:
a(t)=ATh(t-1)+b
z(t)=σ (Wza(t)+Uzh(t-1))
r(t)=σ (Wra(t)+Urh(t-1))
After a large amount of corpus carries out network training, network parameter tends to stablize, so as to form resume thing
Part " knowledge " network.
After forming resume event network, event prediction can be carried out.According to the information to be predicted received, in institute
It states and is predicted in resume event network, and the information of prediction is shown.
It should be noted that volume can receive incoming event information after training above-mentioned resume event " knowledge " network,
It is predicted.It specifically includes: input ve1, ve2..., venIt is context events, obtains the optional sequence v of next eventec1,
vec2..., veck.The relevance function s of front and back event can be defined:
sij=g (Vi (t), Vcj (t))
Relevance function can be replaced with similarity function.Similarity can be defined so that there are as below methods:
Manhattan similarity
Calculate the distance of two vectors:
manhattan(Vi (t), Vcj (t))=∑ | Vi (t)-Vcj (t)|
Cosine display degree
Inner product similarity
Euclidean distance
This patent uses Euclidean distance.After the similarity relationship sij for having context events and candidate events, then
It averages to single candidate events all values, it may be assumed that
Maximum value sj is asked to obtain predicted events all candidate events:
C=maxjsj。
Technical solution of the present invention can automatically extract event chain after inputting resume original text.Resume library based on magnanimity mentions
After taking event, learn and train the resume event network of magnanimity.Have resume event network, can based on existing event into
The prediction of the next event of row, to carry out occupational planning for the talent, to carrying out resume recommendation with robot mechanism, anticipate to the talent
Recommend to work etc..
Third aspect present invention also provides a kind of talents information recommender system, the system comprises: memory, processor and
It is stored in the program for the prediction technique based on reason map that can be run on the memory and on the processor, the base
Such as the above-mentioned prediction side based on reason map is realized when the program of the prediction technique of reason map is executed by the processor
The step of method.
For the present invention by carrying out data extraction to resume inventory, resume data time information is extracted in automation, and is utilized
Deep neural network study, resume event network --- the resume event map for training an event Evolution, further according to
Resume event network carries out the operation such as event prediction, realizes the forecast function to event.It can using technical solution of the present invention
It carries out carrying out talent's occupational planning, carries out resume recommendation to robot mechanism, carry out the offers references such as intention work recommendation to the talent,
User's operation experience and interview success rate are improved, while objective, comprehensive information is provided, reduces screening and analysis
Time and manual read's information the time it takes.
In several embodiments provided herein, it should be understood that disclosed device and method can pass through it
Its mode is realized.Apparatus embodiments described above are merely indicative, for example, the division of the unit, only
A kind of logical function partition, there may be another division manner in actual implementation, such as: multiple units or components can combine, or
It is desirably integrated into another system, or some features can be ignored or not executed.In addition, shown or discussed each composition portion
Mutual coupling or direct-coupling or communication connection is divided to can be through some interfaces, the INDIRECT COUPLING of equipment or unit
Or communication connection, it can be electrical, mechanical or other forms.
Above-mentioned unit as illustrated by the separation member, which can be or may not be, to be physically separated, aobvious as unit
The component shown can be or may not be physical unit；Both it can be located in one place, and may be distributed over multiple network lists
In member；Some or all of units can be selected to achieve the purpose of the solution of this embodiment according to the actual needs.
In addition, each functional unit in various embodiments of the present invention can be fully integrated in one processing unit, it can also
To be each unit individually as a unit, can also be integrated in one unit with two or more units；It is above-mentioned
Integrated unit both can take the form of hardware realization, can also realize in the form of hardware adds SFU software functional unit.
It should be noted that not believing that is be illustrated is the common technology hand of those skilled in the art in the present invention
Section, therefore the present invention is no longer repeated one by one.
Those of ordinary skill in the art will appreciate that: realize that all or part of the steps of above method embodiment can pass through
The relevant hardware of program instruction is completed, and program above-mentioned can store in computer-readable storage medium, which exists
When execution, step including the steps of the foregoing method embodiments is executed；And storage medium above-mentioned includes: movable storage device, read-only deposits
Reservoir (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or
The various media that can store program code such as CD.
If alternatively, the above-mentioned integrated unit of the present invention is realized in the form of software function module and as independent product
When selling or using, it also can store in a computer readable storage medium.Based on this understanding, the present invention is implemented
Substantially the part that contributes to existing technology can be embodied in the form of software products the technical solution of example in other words,
The computer software product is stored in a storage medium, including some instructions are used so that computer equipment (can be with
It is personal computer, server or network equipment etc.) execute all or part of each embodiment the method for the present invention.
And storage medium above-mentioned includes: that movable storage device, ROM, RAM, magnetic or disk etc. are various can store program code
Medium.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.
Claims (10)
1. a kind of prediction technique based on reason map characterized by comprising
Obtain resume inventory；
Data parsing is carried out to the resume data in database；
Event Distillation is carried out to the data after parsing, forms resume event network；
It according to the information to be predicted received, is predicted in the resume event network, and the information of prediction is opened up
Show.
2. the prediction technique according to claim 1 based on reason map, which is characterized in that the data after described pair of parsing
Event Distillation is carried out, resume event network is formed, comprising:
Sentence segmentation is carried out to the data after parsing, forms small sentence unit；
Interdependent syntactic analysis is carried out for the small sentence unit, event information is obtained, forms resume event network.
3. the prediction technique according to claim 2 based on reason map, which is characterized in that described to be directed to the small sentence
Unit carries out interdependent syntactic analysis, obtains event information, comprising:
Based on maximum entropy model and maximum spanning tree model carry out interdependent syntactic analysis or based on conditional random field models carry out according to
Syntactic analysis is deposited, event information is obtained.
4. the prediction technique according to any one of claim 1 to 3 based on reason map, which is characterized in that the shape
At resume event network, comprising:
Word segmentation processing is carried out to event information by participle technique, is then switched to participle according to trained word2vec model
Vector；
The vector of all participles is merged into a vector；
Based on gate figure neural network GGNN algorithm, resume event network is generated.
5. the prediction technique according to claim 4 based on reason map, which is characterized in that it is described by all participles to
Amount merges into a vector, comprising:
Based on one or more of average algorithm, linear transformation algorithm, stitching algorithm, the vector of all participles is merged into
One vector.
6. a kind of prediction meanss based on reason map characterized by comprising
Acquiring unit, for obtaining resume inventory；
Resolution unit, for carrying out data parsing to the resume data in database；
Construction unit forms resume event network for carrying out Event Distillation to the data after parsing；
Predicting unit predicted in the resume event network according to the information to be predicted received, and by the letter of prediction
Breath is shown.
7. the prediction meanss according to claim 6 based on reason map, which is characterized in that the construction unit, comprising:
Cutting unit forms small sentence unit for carrying out sentence segmentation to the data after parsing；
Analytical unit obtains event information, forms resume event for carrying out interdependent syntactic analysis for the small sentence unit
Network.
8. the prediction meanss according to claim 7 based on reason map, which is characterized in that the analytical unit, comprising:
Based on maximum entropy model and maximum spanning tree model carry out interdependent syntactic analysis or based on conditional random field models carry out according to
Syntactic analysis is deposited, event information is obtained.
9. the prediction meanss based on reason map according to any one of claim 6 to 8, which is characterized in that the structure
Build unit, comprising:
Conversion unit, for carrying out word segmentation processing to event information by participle technique, then according to trained word2vec
Participle is switched to vector by model；
Combining unit, for the vector of all participles to be merged into a vector；
Computing unit, for generating resume event network based on gate figure neural network GGNN algorithm.
10. a kind of forecasting system based on reason map, which is characterized in that the system comprises: memory, processor and storage
It is described to be based on thing on the memory and the program of the prediction technique based on reason map that can run on the processor
Manage realization when the program of prediction technique of map is executed by the processor as described in any one of claims 1 to 5 based on
The step of prediction technique of reason map.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811375960.0A CN109460479A (en) | 2018-11-19 | 2018-11-19 | A kind of prediction technique based on reason map, device and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811375960.0A CN109460479A (en) | 2018-11-19 | 2018-11-19 | A kind of prediction technique based on reason map, device and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109460479A true CN109460479A (en) | 2019-03-12 |
Family
ID=65610955
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811375960.0A Pending CN109460479A (en) | 2018-11-19 | 2018-11-19 | A kind of prediction technique based on reason map, device and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109460479A (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110175226A (en) * | 2019-05-09 | 2019-08-27 | 厦门邑通软件科技有限公司 | A kind of dialogue decision-making technique based on various dimensions scene analysis |
| CN110443571A (en) * | 2019-07-16 | 2019-11-12 | 阿里巴巴集团控股有限公司 | The method, device and equipment of knowledge based map progress resume assessment |
| CN110781317A (en) * | 2019-10-29 | 2020-02-11 | 北京明略软件***有限公司 | Method and device for constructing event map and electronic equipment |
| CN110909992A (en) * | 2019-11-11 | 2020-03-24 | 北京明略软件***有限公司 | Risk prediction method, device and equipment |
| CN110968699A (en) * | 2019-11-01 | 2020-04-07 | 数地科技(北京)有限公司 | Logic map construction and early warning method and device based on event recommendation |
| CN111178071A (en) * | 2019-12-26 | 2020-05-19 | 北京明略软件***有限公司 | Method and device for processing resume information and computer readable storage medium |
| CN111339400A (en) * | 2020-02-07 | 2020-06-26 | 北京明略软件***有限公司 | Method, device, computer storage medium and terminal for realizing article recommendation |
| CN111339401A (en) * | 2020-02-07 | 2020-06-26 | 北京明略软件***有限公司 | Method and device for recommending articles, computer storage medium and terminal |
| CN111598239A (en) * | 2020-07-27 | 2020-08-28 | 江苏联著实业股份有限公司 | Method and device for extracting process system of article based on graph neural network |
| CN111737486A (en) * | 2020-05-28 | 2020-10-02 | 广东轩辕网络科技股份有限公司 | Human-sentry matching method and storage device based on knowledge graph and deep learning |
| CN111797233A (en) * | 2020-06-12 | 2020-10-20 | 南京擎盾信息科技有限公司 | Method and device for identifying event chain structure based on vertical field |
| CN111914094A (en) * | 2019-05-10 | 2020-11-10 | 中国人民大学 | Knowledge graph representation learning method based on ternary interaction |
| CN112037181A (en) * | 2020-08-12 | 2020-12-04 | 深圳大学 | 2D SAXS atlas analysis model training method and device |
| CN112612931A (en) * | 2020-12-14 | 2021-04-06 | 国网辽宁省电力有限公司经济技术研究院 | Collaborative filtering recommendation method based on user and expert coupling relation analysis |
| CN113554174A (en) * | 2021-08-11 | 2021-10-26 | 上海明略人工智能(集团)有限公司 | Causal relationship extraction method and system, electronic device and medium |
| CN114398500A (en) * | 2022-01-29 | 2022-04-26 | 哈尔滨工业大学 | Event prediction method based on graph-enhanced pre-training model |
| CN115455205A (en) * | 2022-09-21 | 2022-12-09 | 深圳今日人才信息科技有限公司 | Time sequence knowledge graph-based occupational development planning method |
| CN117171398A (en) * | 2023-11-01 | 2023-12-05 | 浙江大学高端装备研究院 | Method, device and equipment for constructing service tree of industrial Internet platform |
| CN117744786A (en) * | 2024-02-19 | 2024-03-22 | 中国电子科技集团公司第十研究所 | Super-event map network construction and event evolution analysis method |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108764835A (en) * | 2018-05-24 | 2018-11-06 | 广州合摩计算机科技有限公司 | Reverse talent's pushed information method and apparatus |
-
2018
- 2018-11-19 CN CN201811375960.0A patent/CN109460479A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108764835A (en) * | 2018-05-24 | 2018-11-06 | 广州合摩计算机科技有限公司 | Reverse talent's pushed information method and apparatus |
Non-Patent Citations (1)
| Title |
|---|
| ZHONGYANG LI等: "Constructing Narrative Event Evolutionary Graph for Script Event Prediction", 《ARXIV:1805.05081V2 [CS.AI]》 * |
Cited By (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110175226A (en) * | 2019-05-09 | 2019-08-27 | 厦门邑通软件科技有限公司 | A kind of dialogue decision-making technique based on various dimensions scene analysis |
| CN110175226B (en) * | 2019-05-09 | 2021-06-08 | 厦门邑通软件科技有限公司 | Dialogue decision method based on multi-dimensional scene analysis |
| CN111914094A (en) * | 2019-05-10 | 2020-11-10 | 中国人民大学 | Knowledge graph representation learning method based on ternary interaction |
| CN111914094B (en) * | 2019-05-10 | 2023-09-26 | 中国人民大学 | Knowledge graph representation learning method based on ternary interaction |
| CN110443571A (en) * | 2019-07-16 | 2019-11-12 | 阿里巴巴集团控股有限公司 | The method, device and equipment of knowledge based map progress resume assessment |
| CN110781317A (en) * | 2019-10-29 | 2020-02-11 | 北京明略软件***有限公司 | Method and device for constructing event map and electronic equipment |
| CN110781317B (en) * | 2019-10-29 | 2022-03-01 | 北京明略软件***有限公司 | Method and device for constructing event map and electronic equipment |
| CN110968699A (en) * | 2019-11-01 | 2020-04-07 | 数地科技(北京)有限公司 | Logic map construction and early warning method and device based on event recommendation |
| CN110968699B (en) * | 2019-11-01 | 2023-07-18 | 数地工场(南京)科技有限公司 | Logic map construction and early warning method and device based on fact recommendation |
| CN110909992A (en) * | 2019-11-11 | 2020-03-24 | 北京明略软件***有限公司 | Risk prediction method, device and equipment |
| CN111178071A (en) * | 2019-12-26 | 2020-05-19 | 北京明略软件***有限公司 | Method and device for processing resume information and computer readable storage medium |
| CN111339401B (en) * | 2020-02-07 | 2023-05-26 | 北京明略软件***有限公司 | Method and device for recommending articles, computer storage medium and terminal |
| CN111339400B (en) * | 2020-02-07 | 2023-05-23 | 北京明略软件***有限公司 | Method, device, computer storage medium and terminal for realizing article recommendation |
| CN111339400A (en) * | 2020-02-07 | 2020-06-26 | 北京明略软件***有限公司 | Method, device, computer storage medium and terminal for realizing article recommendation |
| CN111339401A (en) * | 2020-02-07 | 2020-06-26 | 北京明略软件***有限公司 | Method and device for recommending articles, computer storage medium and terminal |
| CN111737486A (en) * | 2020-05-28 | 2020-10-02 | 广东轩辕网络科技股份有限公司 | Human-sentry matching method and storage device based on knowledge graph and deep learning |
| CN111737486B (en) * | 2020-05-28 | 2023-06-02 | 广东轩辕网络科技股份有限公司 | Person post matching method and storage device based on knowledge graph and deep learning |
| CN111797233A (en) * | 2020-06-12 | 2020-10-20 | 南京擎盾信息科技有限公司 | Method and device for identifying event chain structure based on vertical field |
| CN111598239B (en) * | 2020-07-27 | 2020-11-06 | 江苏联著实业股份有限公司 | Method and device for extracting process system of article based on graph neural network |
| CN111598239A (en) * | 2020-07-27 | 2020-08-28 | 江苏联著实业股份有限公司 | Method and device for extracting process system of article based on graph neural network |
| CN112037181B (en) * | 2020-08-12 | 2023-09-08 | 深圳大学 | 2D SAXS (three dimensional architecture) atlas analysis model training method and device |
| CN112037181A (en) * | 2020-08-12 | 2020-12-04 | 深圳大学 | 2D SAXS atlas analysis model training method and device |
| CN112612931A (en) * | 2020-12-14 | 2021-04-06 | 国网辽宁省电力有限公司经济技术研究院 | Collaborative filtering recommendation method based on user and expert coupling relation analysis |
| CN113554174A (en) * | 2021-08-11 | 2021-10-26 | 上海明略人工智能(集团)有限公司 | Causal relationship extraction method and system, electronic device and medium |
| CN114398500A (en) * | 2022-01-29 | 2022-04-26 | 哈尔滨工业大学 | Event prediction method based on graph-enhanced pre-training model |
| CN115455205A (en) * | 2022-09-21 | 2022-12-09 | 深圳今日人才信息科技有限公司 | Time sequence knowledge graph-based occupational development planning method |
| CN117171398A (en) * | 2023-11-01 | 2023-12-05 | 浙江大学高端装备研究院 | Method, device and equipment for constructing service tree of industrial Internet platform |
| CN117171398B (en) * | 2023-11-01 | 2024-02-09 | 浙江大学高端装备研究院 | Method, device and equipment for constructing service tree of industrial Internet platform |
| CN117744786A (en) * | 2024-02-19 | 2024-03-22 | 中国电子科技集团公司第十研究所 | Super-event map network construction and event evolution analysis method |
| CN117744786B (en) * | 2024-02-19 | 2024-05-07 | 中国电子科技集团公司第十研究所 | Super logic graph network construction and event evolution analysis method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109460479A (en) | A kind of prediction technique based on reason map, device and system | |
| CN111291185B (en) | Information extraction method, device, electronic equipment and storage medium | |
| CN106202010B (en) | Method and apparatus based on deep neural network building Law Text syntax tree | |
| CN110175227B (en) | Dialogue auxiliary system based on team learning and hierarchical reasoning | |
| CN111078836B (en) | Machine reading understanding method, system and device based on external knowledge enhancement | |
| CN110032632A (en) | Intelligent customer service answering method, device and storage medium based on text similarity | |
| CN110162749A (en) | Information extracting method, device, computer equipment and computer readable storage medium | |
| CN106448670A (en) | Dialogue automatic reply system based on deep learning and reinforcement learning | |
| CN105528437B (en) | A kind of question answering system construction method extracted based on structured text knowledge | |
| CN111222318B (en) | Trigger word recognition method based on double-channel bidirectional LSTM-CRF network | |
| CN111274790A (en) | Chapter-level event embedding method and device based on syntactic dependency graph | |
| CN113821605B (en) | Event extraction method | |
| CN113268609A (en) | Dialog content recommendation method, device, equipment and medium based on knowledge graph | |
| CN110390052A (en) | Search for recommended method, the training method of CTR prediction model, device and equipment | |
| CN115564393A (en) | Recruitment requirement similarity-based job recommendation method | |
| CN111177402B (en) | Evaluation method, device, computer equipment and storage medium based on word segmentation processing | |
| CN114911945A (en) | Knowledge graph-based multi-value chain data management auxiliary decision model construction method | |
| CN115526590A (en) | Efficient human-sentry matching and re-pushing method combining expert knowledge and algorithm | |
| CN113988071A (en) | Intelligent dialogue method and device based on financial knowledge graph and electronic equipment | |
| CN114358657A (en) | Post recommendation method and device based on model fusion | |
| CN115223021A (en) | Visual question-answering-based fruit tree full-growth period farm work decision-making method | |
| Parker et al. | Named entity recognition through deep representation learning and weak supervision | |
| CN114298011B (en) | Neural network, training method, aspect emotion analysis method, device and storage medium | |
| KR20200082136A (en) | Apparatus and method for classfying user's gender identity based on online data | |
| CN115186670A (en) | Method and system for identifying domain named entities based on active learning |
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: 20190312 |