CN115062220A - Attention merging-based recruitment recommendation system - Google Patents
Attention merging-based recruitment recommendation system Download PDFInfo
- Publication number
- CN115062220A CN115062220A CN202210679910.1A CN202210679910A CN115062220A CN 115062220 A CN115062220 A CN 115062220A CN 202210679910 A CN202210679910 A CN 202210679910A CN 115062220 A CN115062220 A CN 115062220A
- Authority
- CN
- China
- Prior art keywords
- user
- recruitment
- module
- attention
- information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Computational Linguistics (AREA)
- Entrepreneurship & Innovation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Economics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a recruitment recommendation system based on attention merging, which comprises a preprocessing module, a recommendation module and a recommendation module, wherein the preprocessing module is used for extracting user information, user recruitment expectation information and recruitment information in original recruitment data; the embedding module is set to respectively obtain a first user characteristic vector, a user recruitment expected characteristic vector and a recruitment characteristic vector; the attention merging module is configured to merge the first user feature vector and the user recruitment expectation feature vector into a second user feature vector; and the prediction output module is used for splicing the second user characteristic vector and the recruitment characteristic vector and then obtaining the call completing rate representing the probability of the call completing and recruitment information of the user. The invention improves the effectiveness of user representation learning in strong correlation field and weak correlation field through the attention merging module; the feature vectors with different lengths are averagely pooled into the feature vectors with the same length through the pooling module, and the problem that long-tail content cannot be effectively recommended due to the Martian effect is solved.
Description
Technical Field
The invention relates to the field of interest recommendation, in particular to an attention merging-based recruitment recommendation system.
Background
At present, technical workers in the field of building and machinery manufacturing in China have high mobility, and the workers generally have a flexible employment form with enterprises or workers, namely, the enterprises or workers are workers engaged for a short time when alive, and the workers are also out for use in leisure time. The demand flexibility of both parties is great, and the conventional recruitment website and system cannot completely meet the demands of both parties at present. In reality, more two people are called and found out by telephone, but due to the asymmetry of information, the two parties cannot know the requirements of the other party before communicating, so that an enterprise or a worker can not find a worker meeting the requirements of the enterprise or the worker, and the worker can not find the work of a heart instrument, the calling efficiency is low, the time of the two parties is wasted invisibly, and the normal work is delayed.
The recommendation of job recruitment based on the call records of worker users is a new type of business that has recently emerged. The enterprise or the job head telephone matched with the expectation of the worker is recommended to the worker by depending on a mature recommendation algorithm, so that accurate recruitment recommendation is realized. However, the current recommendation algorithm has the following problems:
the traditional deep learning technology has the problems of strong feature overfitting and weak feature under-fitting and the Martian effect, so that the effect of the recommendation algorithm based on the deep learning technology is poor. The "martial effect" is a bipolar differentiation phenomenon, that is, most of the existing recommendation algorithms can only provide recommendations for the mainstream hot goods or items, and the probability of recommending the cold goods or items at the tail of the long tail to the user is relatively low. Due to the lack of behavior characteristics, the long-tail content which is interested by the user is difficult to obtain wide and timely recommendation, so that the recommended content is low in quality and poor in diversity.
Overfitting of strong features and under-fitting of weak features are problems with traditional deep learning based recommendation algorithms. The main reason is that they are established on the basis of the Embedding & MLP (multi layer perceptron) paradigm, which ultimately results in poor accuracy of the recommendation result. When an enterprise or a worker carries out recruitment, the overfitting fields are closely related to the interest of most workers in recruitment information, namely, the recommendation result is determined by a few fields representing strong related fields. But some workers have low interest in the overfitting field, so that the model is difficult to effectively recommend to the part of users.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a method which can well solve the problems of over-fitting and under-fitting of strong features and weak features when the characteristics of multiple fields are faced, improves the effectiveness of a system on characterization learning of workers, and solves the problem that long-tail content cannot be effectively recommended due to the Martian effect.
In order to achieve the above object, the present invention provides a technical solution comprising:
the attention merging-based recruitment recommendation system comprises a preprocessing module, an embedding module, an attention merging module and a prediction output module which are sequentially connected;
the preprocessing module is set to extract user information, user recruitment expectation information and recruitment information in original recruitment data;
the embedding module is configured to process the user information, the user recruitment expectation information and the recruitment information into a first user feature vector, a user recruitment expectation feature vector and a recruitment feature vector respectively;
the attention merging module is configured to merge the first user feature vector and a user recruitment desire feature vector into a second user feature vector;
the prediction output module is used for splicing the second user characteristic vector and the recruitment characteristic vector to obtain a dial-through rate representing the probability of the user dialing the recruitment information, and recommending the recruitment information corresponding to the dial-through rate to the user with the highest dial-through rate.
In some preferred embodiments, the user information includes a user id, a user dialing record, a user search record, a user work type, a user location, and a user desired work place;
the user recruitment expected information comprises a user search record, a user work type, a user location and a user expected work place;
the recruitment information comprises recruitment information id, a recruitment information title, release time, work type and the region.
In some preferred embodiments, the prediction output module is further configured to recommend the recruitment information corresponding to the dial-up rate to the other user matched with the user with the highest dial-up rate; the matching comprises: the user work type is the same as the user location.
In some preferred embodiments, a pooling module is further disposed between the embedding module and the attention merging module; the pooling module is configured to average pool feature vectors of different lengths into feature vectors of the same length.
In some preferred embodiments, the attention merging module further comprises a feature domain dividing module for dividing the first user feature vector and the user recruitment expectation feature vector, and assigning an attention weight and an activation unit for the feature domain.
In some preferred embodiments, the method of merging the first user feature vector and the user recruitment expectation feature vector into the second user feature vector comprises:
the first user characteristic vector and the user recruitment expected characteristic vector are connected in series from the second dimension to obtain a matrix E epsilon R n×m (ii) a Wherein m is the dimension of the vectors after the vectors are connected in series, and n is the number of the characteristic domains;
calculating an attention weight vector h:
a＝tanh(W 1 E)
wherein, W 1 ∈R k×n And W 2 ∈R k As a weight matrix, k represents a notationSize of the force unit, a ∈ R n An activation unit representing the characteristic domain, h ∈ R n ，i∈[0,n)，j∈[0,n)；
Merging the attention weight vector h and the matrix E into a second user feature vector Z:
Z＝hE
wherein Z ∈ R m 。
In some preferred embodiments, the objective function of the prediction output module is set as a log-likelihood function:
wherein the content of the first and second substances,
the training set is N, x is the input of the prediction output module, y belongs to {0,1}, and means that when the recruitment information is a positive sample, y is 1, otherwise, y is 0; and p (x) is the output of the prediction output module and represents the probability that the predicted recruitment information is dialed up.
Advantageous effects
The attention merging module distributes reasonable weight and activates different nerve units according to the user characteristics and the recruitment characteristics of the context, so that the effectiveness of user characterization learning in strong related fields and weak related fields is improved; the feature vectors with different lengths are averagely pooled into the feature vectors with the same length through the pooling module, and the problem that long-tail content cannot be effectively recommended due to the Martian effect is solved.
Drawings
FIG. 1 is a schematic system diagram of a preferred embodiment of the present invention;
FIG. 2 is a schematic system configuration according to another preferred embodiment of the present invention;
FIG. 3 is a schematic diagram of data flow and system architecture of another preferred embodiment of the present invention;
FIG. 4 is a graphical illustration of a comparison of AUC and loss results for a preferred embodiment of the invention and a control experiment;
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described with reference to the accompanying drawings. In the description of the present invention, it is to be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", and the like, indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, are merely for convenience in describing the present invention and simplifying the description, and do not indicate or imply that the device or element being referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus, should not be construed as limiting the present invention.
Examples
As shown in fig. 1, the present embodiment discloses an attention merging-based recruitment recommendation system, which includes a preprocessing module, an embedding module, an attention merging module, and a prediction output module, which are connected in sequence;
the preprocessing module is set to extract user information, user recruitment expectation information and recruitment information in original recruitment data; in this embodiment, the original job data mainly includes data provided by a mobile operator, and mainly includes job information, a user dialing record, user work type and area information, a user desired work place, a user search record, a work type index, and an area index, and the following briefly introduces fields included in each information:
the job recruitment information comprises: pid (job information id), pro _ region (area where the job information is located), wlt _ id (job id), create _ time (creation time), title (job information title);
the user call record comprises: uid (user id), type (dial-through type, if the type is 1, the dial-through record is a worker finding record, if the type is 2, the dial-through record is a workhead or enterprise recruitment record, and only the record with the type of 1, object _ id (recruitment information id dialed by the user), call _ time (user dial-through time) are selected herein;
the user work type area information includes: uid (user id), wlt _ id (work kind id), city _ code (city id where the user is located);
the user desires to work including: uid (user id), city _ code (user-expected-work-city id);
the user search record includes: uid (user id), keyword (user search word);
the work category index includes: wlt _ id (job id), wlt _ name (job name);
the region index includes: city _ code (city id), city _ name (city name);
it should be appreciated that, due to the faster database update speed, the enrollment records dialed through by a large percentage of users cannot be found in the enrollment information table, and many users do not search for records and expect to work to record. If the part of users with less information content are directly put into model training, the model effect is affected, so the embodiment needs to screen out the seed users with more information content. If the call records of a certain user can be found in the recruitment information table, the user is selected as the seed user.
The preprocessing module extracts user information, user recruitment expectation information and recruitment information required by the subsequent steps from the original recruitment data; in some preferred embodiments, the user information includes user ID, user dial-up record, user search record, user type, user location, and user desired work place, and for each dial-up record of the user, the dial-up time is recorded, and the user dial-up time is differentiated from the invitation information release time, i.e., call _ time-create _ time, and the record is used to learn the user's preference for new and old invitation information, e.g., { user ID:1721, user dial-up record: [ (2193,17289378973), (8324,1984739084) ], user search record: [ install door, day knot ], user type: 54, customer premises: 370200, the user desires to work: 440100 }; further, in order to further mine the user information so that the subsequent attention combination can reflect more feature domain features, on the basis of the user information, user recruitment expected information including user search records, user work types, user locations and user expected work places is extracted; the job invitation information includes a job invitation information ID, a job invitation information title, a release time, a job type, and a region where, for example, { job invitation information ID: 7354, information title: "economic development area project installer/air conditioner/duct/ventilation", release time: 1641282268, work type: 375, region: 21574264}.
The embedding module is configured to process the user information, the user recruitment expectation information and the recruitment information into a first user feature vector, a user recruitment expectation feature vector and a recruitment feature vector respectively; in some preferred embodiments, a Bert-base-Chinese version of the pretrained Bert model is used as an embedded module, and is put into a model to train adjustment parameters in order to adapt to a job-taking data set. It should be understood that, since different users have different numbers of call-through records and search records, the number of features included in the user feature vector is different in different instances, which results in different lengths of the user feature vector formed by concatenating a plurality of features. While the subsequent full-connection network can only process input vectors with fixed length, therefore, in some preferred embodiments, it is considered to convert feature vectors with different lengths into feature vectors with the same length by using pooling operation, thereby ensuring that data can be processed by the full-connection layer. Specifically, as shown in fig. 2, a pooling module is further disposed between the embedding module and the attention merging module; the pooling module is configured to average pool feature vectors of different lengths into feature vectors of the same length.
The attention merging module is configured to merge the first user feature vector and a user recruitment desire feature vector into a second user feature vector; it should be understood that the conventional recommendation algorithm is based on a model of the series of layers. The series layer connects different characteristic domains in series to complete parameter learning. During the training process, there are some specific domain overfitting problems, and these overfitting domains are closely related to the interests of most users in the recruitment information, i.e. the recommendation result is determined by a few fields representing strong related domains. But some users have low interest in the over-fitting field, so that the model is difficult to effectively recommend to the users. According to the invention, the attention merging module replaces a series layer in the traditional algorithm, and the user characteristics and the user recruitment expected characteristics are merged together to be used as more comprehensive user representation. In some preferred embodiments, the attention merging module further includes a feature domain partitioning module for partitioning the first user feature vector and the user recruitment expectation feature vector, and assigning an attention weight and an activation unit for the feature domain, where the specific method includes:
the first user characteristic vector and the user recruitment expected characteristic vector are connected in series from the second dimension to obtain a matrix E epsilon R n×m (ii) a Wherein m is the dimension of the vectors after the vectors are connected in series, n is the number of the characteristic domains, and R is a matrix; it should be understood that the first row of a two-dimensional vector (i.e., the first dimension) represents a feature vector, and that stitching from the first dimension is meaningless, and thus stitching from the second dimension is a concatenation of all feature vectors of a user.
Calculating an attention weight vector h:
a＝tanh(W 1 E)；
wherein, W 1 ∈R k×n And W 2 ∈R k For the weight matrix, k represents the size of the attention unit, a ∈ R n An activation unit representing the characteristic domain, h ∈ R n I belongs to [0, n), j belongs to [0, n); the above formula is a softmax formula, and the purpose is normalization, i represents the ith feature domain, and j on the denominator needs to traverse i ∈ [0, n) all the time. It should be understood that the weight matrix W 1 And W 2 The matrix is initialized randomly when training begins, and the optimal weight matrix is obtained by automatic adjustment along with the training process. And the activation unit learns the weight of each feature domain in different user samples in the training process and continuously and automatically adjusts in the training process to obtain the optimal activation unit.
Merging the attention weight vector h and the matrix E into a second user feature vector Z:
z ═ hE; wherein Z ∈ R m 。
The prediction output module is used for splicing the second user characteristic vector and the recruitment characteristic vector to obtain a dial-through rate representing the probability of the user dialing the recruitment information, and recommending the recruitment information corresponding to the dial-through rate to the user with the highest dial-through rate. It should be understood that in this embodiment, the splicing of the second user feature vector and the recruit feature vector is performed by splice of a continate layer.
On the other hand, as mentioned above, because there are users whose information is not enough in the original recruitment data, in some preferred embodiments for this part of users, the prediction output module is further configured to recommend recruitment information corresponding to the dial-up rate to other users matching the user with the highest dial-up rate; the matching comprises: the user work type is the same as the user location.
In some preferred embodiments, a loss function is used to determine whether training of the recommendation system is completed, specifically, the objective function is set as a log-likelihood function:
wherein the content of the first and second substances,
the training set is N, x is the input of the prediction output module, y belongs to {0,1}, and means that when the recruitment information is a positive sample, y is 1, otherwise, y is 0; and p (x) is the output of the prediction output module and represents the probability that the predicted recruitment information is dialed up.
Comparative example
In the comparison test, a Bert-base-Chinese version Chinese pre-training model is adopted as the Bert model of the embedded module, the hidden layer size (hidden size) is 768, the training sample size (batch size) is 4, the learning rate (learning rate) is 1e-5, and the SGD is used as a model optimizer. The used activation function is ReLU, the used learning framework is Pythrch, and the experiment operation platform is a computer with IntelXeon @ E5-2678v3cpu, GTX2080TI video memory of 11G and memory size of 32G. The Attention merging-based recruitment Recommendation system disclosed by the invention is named as RecRecRec (Recirculation Model based on Attention Merging), and the data circulation and the system structure are shown in FIG. 3.
The comparison algorithm system comprises:
LR: logistic Regression (LR) is a shallow model widely used before deep networks for CTR prediction tasks.
YouTube DNN: YouTube DNN follows the Embedding & MLP architecture and is the basis for most of the later developed deep networks for CTR modeling. It provides a strong baseline for our model comparisons.
Wide & Deep: in practical industrial applications, the Wide & Deep model has been widely accepted. It consists of two parts: 1) The Wide model is used for processing the characteristics of the manually designed cross products; 2) and the Deep model automatically extracts the nonlinear relation among the characteristics.
PNN: the PNN introduces a product layer after the embedding layer to capture the interaction of the higher order features.
Deep FM: the Deep FM takes a factorization machine as a Wide module on the basis of Wide & Deep models so as to save characteristic engineering operation.
The evaluation indexes include:
AUC is an index widely used in the field of CTR prediction. AUC is the area under the ROC curve, representing the probability that the prediction score for a positive sample is greater than the prediction score for a negative sample. Specifically, if M represents the number of positive samples and N represents the number of negative samples, how many sets of positive samples score more than negative samples scores among all M × N positive and negative sample pairs are calculated and then divided by MN. The time complexity of the calculation method is O (n) 2 ) Where N is M + N, the AUC calculation formula is as follows:
therein, pred pos >pred neg Representing a positive and negative sample pair, the value is 1 if the positive sample score is greater than the negative sample score, otherwise the value is 0.
In addition, the RelaImpr index was introduced to measure the relative improvement to the model. The AUC value for a random guesser was 0.5. As the basis of the deep network of the CTR modeling, a YouTube DNN model is selected as a base model (base model). Therefore, RelaImpr is defined as follows:
results of the experiment
All experiments were repeated 5 times and the average results are reported as shown in table 1. It is clear that all deep networks clearly defeat the LR system, which does justify the strength of deep learning. PNN and Deep FM with specially designed structures perform better than Wide & Deep. The recrecrec proposed by the present invention performed best in all comparison systems.
TABLE 1 comparison of system forecasted results on the Inquiry data set
Comparison of attention merging layer and attention tandem layer
As shown in fig. 4, by replacing the attention-merging layer in the model proposed by the present invention with the conventional attention-tandem layer as a control experimental example, it can be seen that the attention-merging layer performs better than the tandem layer in AUC and loss of the test set. When the feature vectors pass through the cascade layer and feed forward, the same part of the neurons of all users is activated. When the attention merging layer is passed, different neurons will be activated facing different user features, which means that the interaction between different features is learned, which improves the performance of the system.
The foregoing shows and describes the general principles, essential features, and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are given by way of illustration of the principles of the present invention, but that various changes and modifications may be made without departing from the spirit and scope of the invention, and such changes and modifications are within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (7)
1. Attention merging-based recruitment recommendation system is characterized in that: the system comprises a preprocessing module, an embedding module, an attention merging module and a prediction output module which are connected in sequence;
the preprocessing module is set to extract user information, user recruitment expectation information and recruitment information in original recruitment data;
the embedding module is configured to process the user information, the user recruitment expectation information and the recruitment information into a first user feature vector, a user recruitment expectation feature vector and a recruitment feature vector respectively;
the attention merging module is configured to merge the first user feature vector and a user recruitment desire feature vector into a second user feature vector;
the prediction output module is used for splicing the second user characteristic vector and the recruitment characteristic vector to obtain a dial-through rate representing the probability of the user dialing the recruitment information, and recommending the recruitment information corresponding to the dial-through rate to the user with the highest dial-through rate.
2. The attention-merging-based recruitment recommendation system of claim 1, wherein: the user information comprises a user id, a user dialing record, a user searching record, a user work type, a user location and a user expected work place;
the user recruitment expected information comprises a user search record, a user work type, a user location and a user expected work place;
the recruitment information comprises recruitment information id, a recruitment information title, release time, work types and the region.
3. The attention-merging-based recruitment recommendation system of claim 2, wherein: the prediction output module is also set to recommend recruitment information corresponding to the dial-up rate to other users matched with the user with the highest dial-up rate; the matching comprises: the user work type is the same as the user location.
4. The attention-merging-based recruitment recommendation system of claim 1, wherein: a pooling module is also arranged between the embedding module and the attention merging module; the pooling module is configured to average pool feature vectors of different lengths into feature vectors of the same length.
5. The attention-merging-based recruitment recommendation system of claim 1, wherein: the attention merging module also comprises a feature domain division unit which divides the first user feature vector and the user recruitment expected feature vector respectively, and allocates attention weight and an activation unit for the feature domain.
6. The attention merging-based recruitment recommendation system of claim 5 wherein: the method for merging the first user feature vector and the user recruitment expectation feature vector into the second user feature vector comprises the following steps:
the first user characteristic vector and the user recruitment expected characteristic vector are connected in series from the second dimension to obtain a matrix E epsilon R n ×m (ii) a Wherein m is the dimension of the vectors after the vectors are connected in series, and n is the number of the characteristic domains;
calculating an attention weight vector h:
a＝tanh(W 1 E)
wherein, W 1 ∈R k×n And W 2 ∈R k For the weight matrix, k represents the size of the attention unit, a ∈ R n An activation unit representing the characteristic domain, h ∈ R n ，i∈[0,n)，j∈[0,n)；
Merging the attention weight vector h and the matrix E into a second user feature vector Z:
Z＝hE
wherein Z ∈ R m 。
7. The attention-merging-based recruitment recommendation system of claim 1, wherein: the target function of the prediction output module is set as a log-likelihood function:
wherein the content of the first and second substances,
the training set is N, x is the input of the prediction output module, y belongs to {0,1}, and means that when the recruitment information is a positive sample, y is 1, otherwise, y is 0; and p (x) is the output of the prediction output module and represents the probability that the predicted recruitment information is dialed up.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210679910.1A CN115062220B (en) | 2022-06-16 | 2022-06-16 | Attention merging-based recruitment recommendation system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210679910.1A CN115062220B (en) | 2022-06-16 | 2022-06-16 | Attention merging-based recruitment recommendation system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115062220A true CN115062220A (en) | 2022-09-16 |
| CN115062220B CN115062220B (en) | 2023-06-23 |
Family
ID=83200083
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210679910.1A Active CN115062220B (en) | 2022-06-16 | 2022-06-16 | Attention merging-based recruitment recommendation system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115062220B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115481220A (en) * | 2022-09-20 | 2022-12-16 | 深圳今日人才信息科技有限公司 | Post and resume content-based intelligent matching method and system for comparison learning human posts |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105488662A (en) * | 2016-01-07 | 2016-04-13 | 北京歌利沃夫企业管理有限公司 | Bi-directional recommendation-based online recruitment system |
| CN107578214A (en) * | 2017-09-05 | 2018-01-12 | 四川民工加网络科技有限公司 | A kind of method and system of building trade recruitment |
| CN107679757A (en) * | 2017-09-30 | 2018-02-09 | 四川民工加网络科技有限公司 | The matching process and device of services dispatch |
| CN108960787A (en) * | 2018-08-14 | 2018-12-07 | 安徽网才信息技术股份有限公司 | A method of it is issued for enterprise and updates recruitment casual labour's information |
| US20190114593A1 (en) * | 2017-10-17 | 2019-04-18 | ExpertHiring, LLC | Method and system for managing, matching, and sourcing employment candidates in a recruitment campaign |
| CN109949012A (en) * | 2019-03-25 | 2019-06-28 | 贵州爱唐文化网络科技有限公司 | It works in short term on a kind of line the method and interaction platform of process dynamic interaction |
| CN109960759A (en) * | 2019-03-22 | 2019-07-02 | 中山大学 | Recommender system clicking rate prediction technique based on deep neural network |
| CN111062775A (en) * | 2019-12-03 | 2020-04-24 | 中山大学 | Recommendation system recall method based on attention mechanism |
| CN111339415A (en) * | 2020-02-25 | 2020-06-26 | 中国科学技术大学 | Click rate prediction method and device based on multi-interactive attention network |
| KR20210047284A (en) * | 2020-05-08 | 2021-04-29 | 바이두 온라인 네트웍 테크놀러지 (베이징) 캄파니 리미티드 | Recruitment position description text generation method, device, apparatus and medium |
| CN113641904A (en) * | 2021-08-16 | 2021-11-12 | 上海花千树信息科技有限公司 | Method and device for recommending target customers for stores under Internet marriage line |
| CN114169869A (en) * | 2022-02-14 | 2022-03-11 | 北京大学 | Attention mechanism-based post recommendation method and device |
| CN114358657A (en) * | 2022-03-09 | 2022-04-15 | 北京大学 | Post recommendation method and device based on model fusion |
-
2022
- 2022-06-16 CN CN202210679910.1A patent/CN115062220B/en active Active
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105488662A (en) * | 2016-01-07 | 2016-04-13 | 北京歌利沃夫企业管理有限公司 | Bi-directional recommendation-based online recruitment system |
| CN107578214A (en) * | 2017-09-05 | 2018-01-12 | 四川民工加网络科技有限公司 | A kind of method and system of building trade recruitment |
| CN107679757A (en) * | 2017-09-30 | 2018-02-09 | 四川民工加网络科技有限公司 | The matching process and device of services dispatch |
| US20190114593A1 (en) * | 2017-10-17 | 2019-04-18 | ExpertHiring, LLC | Method and system for managing, matching, and sourcing employment candidates in a recruitment campaign |
| CN108960787A (en) * | 2018-08-14 | 2018-12-07 | 安徽网才信息技术股份有限公司 | A method of it is issued for enterprise and updates recruitment casual labour's information |
| CN109960759A (en) * | 2019-03-22 | 2019-07-02 | 中山大学 | Recommender system clicking rate prediction technique based on deep neural network |
| CN109949012A (en) * | 2019-03-25 | 2019-06-28 | 贵州爱唐文化网络科技有限公司 | It works in short term on a kind of line the method and interaction platform of process dynamic interaction |
| CN111062775A (en) * | 2019-12-03 | 2020-04-24 | 中山大学 | Recommendation system recall method based on attention mechanism |
| CN111339415A (en) * | 2020-02-25 | 2020-06-26 | 中国科学技术大学 | Click rate prediction method and device based on multi-interactive attention network |
| KR20210047284A (en) * | 2020-05-08 | 2021-04-29 | 바이두 온라인 네트웍 테크놀러지 (베이징) 캄파니 리미티드 | Recruitment position description text generation method, device, apparatus and medium |
| CN113627135A (en) * | 2020-05-08 | 2021-11-09 | 百度在线网络技术（北京）有限公司 | Method, device, equipment and medium for generating recruitment post description text |
| CN113641904A (en) * | 2021-08-16 | 2021-11-12 | 上海花千树信息科技有限公司 | Method and device for recommending target customers for stores under Internet marriage line |
| CN114169869A (en) * | 2022-02-14 | 2022-03-11 | 北京大学 | Attention mechanism-based post recommendation method and device |
| CN114358657A (en) * | 2022-03-09 | 2022-04-15 | 北京大学 | Post recommendation method and device based on model fusion |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115481220A (en) * | 2022-09-20 | 2022-12-16 | 深圳今日人才信息科技有限公司 | Post and resume content-based intelligent matching method and system for comparison learning human posts |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115062220B (en) | 2023-06-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109213863B (en) | Learning style-based adaptive recommendation method and system | |
| CN108509573B (en) | Book recommendation method and system based on matrix decomposition collaborative filtering algorithm | |
| CN109087178A (en) | Method of Commodity Recommendation and device | |
| CN107292097B (en) | Chinese medicine principal symptom selection method based on feature group | |
| US20210279929A1 (en) | Training and using an ensemble of complimentary convolutional neural networks for cross-domain retrieval of fashion item images | |
| CN108427740B (en) | Image emotion classification and retrieval algorithm based on depth metric learning | |
| KR101992988B1 (en) | An online shopping mall system recommending apparel materials using dynamic learning method | |
| CN108763367B (en) | Method for recommending academic papers based on deep alignment matrix decomposition model | |
| CN106776869A (en) | Chess game optimization method, device and search engine based on neutral net | |
| CN111563770A (en) | Click rate estimation method based on feature differentiation learning | |
| CN112632296B (en) | Knowledge graph-based paper recommendation method and system with interpretability and terminal | |
| CN113159171A (en) | Plant leaf image fine classification method based on counterstudy | |
| CN116127186A (en) | Knowledge-graph-based individual matching recommendation method and system for person sentry | |
| CN115062220A (en) | Attention merging-based recruitment recommendation system | |
| CN116128254A (en) | Embedded intelligent manufacturing resource configuration method and terminal system based on edge calculation | |
| He et al. | Dadnn: multi-scene ctr prediction via domain-aware deep neural network | |
| CN110378269A (en) | Pass through the movable method not previewed in image query positioning video | |
| CN114840745A (en) | Personalized recommendation method and system based on graph feature learning and deep semantic matching model | |
| CN114781503A (en) | Click rate estimation method based on depth feature fusion | |
| CN110175631A (en) | A kind of multiple view clustering method based on common Learning Subspaces structure and cluster oriental matrix | |
| CN111583363B (en) | Visual automatic generation method and system for graphic news | |
| CN116383422B (en) | Non-supervision cross-modal hash retrieval method based on anchor points | |
| Wang et al. | Multi‐feedback Pairwise Ranking via Adversarial Training for Recommender | |
| CN115982373A (en) | Knowledge graph recommendation method combining multi-level interactive contrast learning | |
| CN114862511A (en) | Short video live broadcast marketing task recommendation method based on deep 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |