CN113486240A - Position recommendation method based on SWPEM routing algorithm - Google Patents
Position recommendation method based on SWPEM routing algorithm Download PDFInfo
- Publication number
- CN113486240A CN113486240A CN202110770959.3A CN202110770959A CN113486240A CN 113486240 A CN113486240 A CN 113486240A CN 202110770959 A CN202110770959 A CN 202110770959A CN 113486240 A CN113486240 A CN 113486240A
- Authority
- CN
- China
- Prior art keywords
- user
- attention
- dimensional
- clustering
- capsules
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 16
- 239000013598 vector Substances 0.000 claims abstract description 34
- 239000002775 capsule Substances 0.000 claims abstract description 24
- 230000007246 mechanism Effects 0.000 claims abstract description 24
- 238000012549 training Methods 0.000 claims description 3
- 238000000605 extraction Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000006399 behavior Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000007115 recruitment Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000001939 inductive effect Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000011524 similarity measure Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—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
- 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
- G06F16/906—Clustering; Classification
-
- 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)
- Databases & Information Systems (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- Strategic Management (AREA)
- General Engineering & Computer Science (AREA)
- Operations Research (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- Marketing (AREA)
- Economics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a position recommendation method based on an SWP EM routing algorithm, which comprises the following steps: coding high-dimensional sparse features corresponding to the user information and the job browsing history data, and converting the high-dimensional sparse features into low-dimensional dense vectors; inputting the low-dimensional dense vectors into a feature clustering layer, and clustering the position features in the low-dimensional dense vectors by using GMM to obtain a plurality of clustering clusters; calculating the similarity between the clustering cluster vector and the position vector, and distributing attention weight to different candidate positions according to the similarity; arranging in a reverse order according to the attention weight of the user to each position, and extracting the expected capsules of different positions of the user by using an attention mechanism; and outputting actual recommended positions converted by the plurality of expected capsules by using a softmax function, and returning recommended position combinations corresponding to the maximum value of the attention weights of a plurality of clustering clusters ranked in front by the user to the user. By applying the embodiment of the invention, the problem of weight distribution of overlong job browsing history sequences disturbing attention mechanism is solved.
Description
Technical Field
The invention relates to the technical field of deep learning, in particular to a position recommendation method based on an SWP EM routing algorithm.
Background
The rapid development of the internet enables people to enter an era with complicated information, for example, under the scene of human resource recruitment requirement, the recruitment information is thousands of, and the requirement of an interviewer on the position information becomes diversified and individualized, that is, the interviewer may not be able to search out valuable positions from massive position information. The personalized recommendation technology is developed in order to better provide services for users and assist users to find out the positions really meeting the self expectations faster.
For the research of recommendation technology, early recommendation systems are based on collaborative filtering, and are gradually developed to be based on machine learning, and finally, deep learning formally enters the field of recommendation systems. Currently, the mainstream recommendation method is to use an attention mechanism to extract recommendations that different positions of a user are expected to use for position combination or sequence. The positive effect of attention on the recommended task has been well documented. However, as the tasks that the recommendation models need to handle become more diverse and personalized, traditional attention mechanisms have been unable to meet the needs of the recommendation models.
At present, the recommendation model fusing the traditional attention mechanism mainly faces the following inherent problems:
1. lack of multi-target extraction: the extraction of the expectations of different positions of a user is important for reasonable recommendation, for example, in resume delivery, an interviewer can interview multiple companies in order to increase the probability of taking a recording notice by himself, delivery can be carried out on multiple related positions of the position expected by himself, the expression capacity of a single position embedded vector is insufficient, unreasonable recommendation is often given, and therefore the expectations of different positions of the user need to be represented by multiple vectors.
2. Distraction under a sequence of very long behaviors: many job positions are related, and it is generally considered that the longer the job browsing history sequence of the user is, the more clearly the job position expectation of the user can be described. However, an excessively long job browsing history sequence may disturb the weight assignment of attention mechanism, and may give a higher attention weight to an undesired job, affecting the accuracy of job recommendation.
3. Feature confusion at high dimensions: using high-dimensional vectors to abstractly represent the user's job expectations, it is difficult to capture similarities between jobs, resulting in job signatures that are too generalized and thus lose focus.
Disclosure of Invention
Compared with the prior art, the position recommendation method based on the SWP EM routing algorithm can respectively recommend different expected positions, and under the condition of multi-target extraction, the position recommendation method mechanism combining the SWP EM routing algorithm improves the recommendation accuracy and diversity, solves the problem of excessive user quantity, and improves the user experience.
In order to achieve the above object, the present invention provides a job recommendation method based on SWP EM routing algorithm, including:
coding high-dimensional sparse features corresponding to the user information and the job browsing history data to convert the high-dimensional sparse features into low-dimensional dense vectors;
inputting the low-dimensional dense vector into a feature clustering layer, and clustering the position features in the low-dimensional dense vector by using GMM to obtain a plurality of clustering clusters;
the similarity between the clustering cluster vectors and the position vectors is calculated in parallel through an SWP EM routing algorithm, and attention weights are distributed to different candidate positions according to the similarity;
arranging in a reverse order according to the attention weight of the user to each position, and extracting capsules expected by different positions of the user by using an attention mechanism;
and outputting actual recommended positions converted by the plurality of expected capsules by using a softmax function, and returning recommended position combinations corresponding to the maximum value of the attention weights of a plurality of clustering clusters ranked in front by the user to the user.
Optionally, the step of encoding the high-dimensional sparse features corresponding to the user information and the job browsing history data to convert the high-dimensional sparse features into the low-dimensional dense vectors includes:
and preliminarily encoding the high-dimensional sparse features into low-dimensional dense vectors by using a Doc2Vec algorithm, wherein the high-dimensional sparse features comprise user information at least comprising gender, age and occupation and a job browsing history sequence at least comprising job types, monthly salaries, academic requirements and company properties.
In one implementation, the step of arranging in reverse order according to the attention weight of the user for each position and extracting the capsules expected by the user at different positions by using the attention mechanism includes:
arranging the capsules in a reverse order according to the attention weight of the user to each position, and extracting the capsules expected by different positions of the user by using an attention mechanism;
if the desired features are not included in the training data, the scores are further parameterized using an attention mechanism and weighted summed to obtain the final desired capsule.
By applying the position recommendation method based on the SWP EM routing algorithm, provided by the embodiment of the invention, the high-dimensional characteristics of the browsing history sequence of the user can be fully mined, and an improved attention mechanism is integrated to separate a plurality of position expectations of the user; the similarity among the positions can be captured, the problem of weight distribution that overlong position browsing history sequences disturb attention mechanisms can be solved, and higher attention weight is given to the undesired positions in balance.
Drawings
Fig. 1 is a schematic flow chart of a position recommendation method based on the SWP EM routing algorithm according to an embodiment of the present invention.
Detailed Description
The embodiments of the present invention are described below with reference to specific embodiments, and other advantages and effects of the present invention will be easily understood by those skilled in the art from the disclosure of the present specification. The invention is capable of other and different embodiments and of being practiced or of being carried out in various ways, and its several details are capable of modification in various respects, all without departing from the spirit and scope of the present invention.
As shown in fig. 1, the present invention provides a job recommendation method based on SWP EM routing algorithm, which includes:
s110, coding high-dimensional sparse features corresponding to the user information and the job browsing history data to convert the high-dimensional sparse features into low-dimensional dense vectors;
it should be noted that 12 items of user information such as gender, age, occupation, and the like, and 18 items of job browsing history sequences such as month salary, academic requirement, company property, and the like, which are high-dimensional sparse features, are primarily encoded in the feature encoding layer by the Doc2Vec algorithm, so that the high-dimensional sparse features are converted into a low-dimensional dense vector which can be calculated and structured to be used as input of the feature clustering layer.
S120, inputting the low-dimensional dense vector into a feature clustering layer, and clustering the position features in the low-dimensional dense vector by using GMM to obtain a plurality of clustering clusters;
it can be understood that after the obtained low-dimensional dense vectors are input into the feature clustering layer, different position features in the information are clustered and encoded by using the GMM, and the position features are grouped into a plurality of clustering clusters.
S130, calculating the similarity between the clustering cluster vectors and the position vectors in parallel through an SWP EM routing algorithm, and distributing attention weights to different candidate positions according to the similarity;
it should be noted that, the similarity in the present invention refers to the attention value, and calculating the similarity is an important process for obtaining the attention weight. And calculating the similarity between the clustering cluster vector and the position vector, namely calculating an attention value, and then normalizing the calculated original value to obtain an attention weight. The attention mechanism is a weight parameter allocation mechanism, and the goal is to assist the model in capturing important information. Similarity measures are used to determine the degree of equality between two job vectors, and computing similarity can better understand the degree of closeness between jobs. Attention should be paid to similar positions that are worth recommendation.
Specifically, according to the obtained multiple clustering clusters, similarity between clustering cluster vectors and position vectors is calculated in parallel through an SWP EM routing algorithm, the calculated attention value is normalized to obtain an attention weight, then the attention weights are distributed to different candidate positions, the SWP EM routing algorithm is handed to a Spark distributed computing platform to accelerate calculation, for recommendation services which are updated once every day, a recommendation result is calculated in advance through Spark, and the recommendation result is stored for front-end service calling. And adding a matrix shared attention mechanism (sharing all parameters of a weight matrix) in the original EM routing algorithm to replace the original full-connection attention mechanism, processing the position browsing history sequence with variable length of the user and mapping all high-level capsules to a feature space with the same dimension.
S140, arranging in a reverse order according to the attention weight of the user to each position, and extracting the expected capsules of different positions of the user by using an attention mechanism;
it can be understood that the capsules are arranged in the inductive evaluation layer in a reverse order according to the attention weight of the user to each position, and the expected capsules of different positions of the user are extracted by using an attention mechanism; for a desired feature that never appears in the training data, the attention score for its interaction cannot be estimated, and the present invention further parameterizes the score using the attention mechanism and then performs a weighted summation to arrive at the final desired capsule.
According to the method, the problem that a user browses historical sequences in an indefinite length position and maps all high-level capsules to the feature space of the same dimension is solved, the significance degree of position features is represented by information entropy, and the mutual relation between behavior capsules and expected capsules is fully explained.
And S150, outputting actual recommended positions converted by the multiple expected capsules by using a softmax function, and returning recommended position combinations corresponding to the maximum value of the attention weights of the multiple clustering clusters ranked in front of the user to the user.
It should be noted that the invention uses the softmax function to output the actual recommended positions converted from a plurality of expected capsules, and returns the recommended position combination corresponding to the maximum value of the attention weights of the Top N clustering clusters of the user to the user, thereby realizing recommending the positions to the user, improving the accuracy and diversity of recommendation, solving the problem of too large amount of users, and improving the user experience.
The foregoing embodiments are merely illustrative of the principles and utilities of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or change the above-mentioned embodiments without departing from the spirit and scope of the present invention. Accordingly, it is intended that all equivalent modifications or changes which can be made by those skilled in the art without departing from the spirit and technical spirit of the present invention be covered by the claims of the present invention.
Claims (3)
1. A position recommendation method based on SWP EM routing algorithm is characterized by comprising the following steps:
coding high-dimensional sparse features corresponding to the user information and the job browsing history data to convert the high-dimensional sparse features into low-dimensional dense vectors;
inputting the low-dimensional dense vector into a feature clustering layer, and clustering the position features in the low-dimensional dense vector by using GMM to obtain a plurality of clustering clusters;
the similarity between the clustering cluster vectors and the position vectors is calculated in parallel through an SWP EM routing algorithm, and attention weights are distributed to different candidate positions according to the similarity;
arranging in a reverse order according to the attention weight of the user to each position, and extracting capsules expected by different positions of the user by using an attention mechanism;
and outputting actual recommended positions converted by the plurality of expected capsules by using a softmax function, and returning recommended position combinations corresponding to the maximum value of the attention weights of a plurality of clustering clusters ranked in front by the user to the user.
2. The method for recommending positions based on the SWP EM routing algorithm according to claim 1, wherein the step of encoding the high-dimensional sparse features corresponding to the user information and the position browsing history data to convert them into low-dimensional dense vectors comprises:
and preliminarily encoding the high-dimensional sparse features into low-dimensional dense vectors by using a Doc2Vec algorithm, wherein the high-dimensional sparse features comprise user information at least comprising gender, age and occupation and a job browsing history sequence at least comprising job types, monthly salaries, academic requirements and company properties.
3. The method for position recommendation based on SWP EM routing algorithm as claimed in claim 1, wherein the step of arranging in reverse order according to the attention weight of the user for each position and extracting the capsules expected by the user in different positions by using the attention mechanism comprises:
arranging the capsules in a reverse order according to the attention weight of the user to each position, and extracting the capsules expected by different positions of the user by using an attention mechanism;
if the desired features are not included in the training data, the scores are further parameterized using an attention mechanism and weighted summed to obtain the final desired capsule.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110770959.3A CN113486240B (en) | 2021-07-08 | 2021-07-08 | Position recommendation method based on SWPEM routing algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110770959.3A CN113486240B (en) | 2021-07-08 | 2021-07-08 | Position recommendation method based on SWPEM routing algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113486240A true CN113486240A (en) | 2021-10-08 |
| CN113486240B CN113486240B (en) | 2024-02-02 |
Family
ID=77937446
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110770959.3A Active CN113486240B (en) | 2021-07-08 | 2021-07-08 | Position recommendation method based on SWPEM routing algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113486240B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102184364A (en) * | 2011-05-26 | 2011-09-14 | 南京财经大学 | Semi-supervised learning-based recommendation system shilling attack detection method |
| US8364559B1 (en) * | 2010-01-07 | 2013-01-29 | Amazon Technologies, Inc. | Method, medium, and system of recommending a substitute item |
| CN112364236A (en) * | 2019-07-24 | 2021-02-12 | 阿里巴巴集团控股有限公司 | Target object recommendation system, method and device, and data processing method and device |
| CN112699215A (en) * | 2020-12-24 | 2021-04-23 | 齐鲁工业大学 | Grading prediction method and system based on capsule network and interactive attention mechanism |
| CN112765461A (en) * | 2021-01-12 | 2021-05-07 | 中国计量大学 | Session recommendation method based on multi-interest capsule network |
-
2021
- 2021-07-08 CN CN202110770959.3A patent/CN113486240B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8364559B1 (en) * | 2010-01-07 | 2013-01-29 | Amazon Technologies, Inc. | Method, medium, and system of recommending a substitute item |
| CN102184364A (en) * | 2011-05-26 | 2011-09-14 | 南京财经大学 | Semi-supervised learning-based recommendation system shilling attack detection method |
| CN112364236A (en) * | 2019-07-24 | 2021-02-12 | 阿里巴巴集团控股有限公司 | Target object recommendation system, method and device, and data processing method and device |
| CN112699215A (en) * | 2020-12-24 | 2021-04-23 | 齐鲁工业大学 | Grading prediction method and system based on capsule network and interactive attention mechanism |
| CN112765461A (en) * | 2021-01-12 | 2021-05-07 | 中国计量大学 | Session recommendation method based on multi-interest capsule network |
Non-Patent Citations (3)
| Title |
|---|
| SASHANK SRIDHAR 等: "Twitter Social Networking Graph using Hypernym based Semantic Similarity Detection", 《2020 INTERNATIONAL CONFERENCE ON SMART ELECTRONICS AND COMMUNICATION (ICOSEC)》, pages 28 - 35 * |
| 常思维: "一种基于EM路由算法的推荐模型", 《中国优秀硕士学位论文全文数据库信息科技辑》, no. 4, pages 138 - 1320 * |
| 文益民 等: "基于深度学习的中文网络招聘文本中的技能词抽取方法", 《桂林电子科技大学学报》, vol. 40, no. 4, pages 338 - 348 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113486240B (en) | 2024-02-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112785397B (en) | Product recommendation method, device and storage medium | |
| Wang et al. | Kvt: k-nn attention for boosting vision transformers | |
| Khan et al. | Multi-gcn: Graph convolutional networks for multi-view networks, with applications to global poverty | |
| CN109493199A (en) | Products Show method, apparatus, computer equipment and storage medium | |
| CN107066616A (en) | Method, device and electronic equipment for account processing | |
| CN107545276B (en) | Multi-view learning method combining low-rank representation and sparse regression | |
| CN108710609A (en) | A kind of analysis method of social platform user information based on multi-feature fusion | |
| Maire et al. | Reconstructive sparse code transfer for contour detection and semantic labeling | |
| Ma et al. | Spatial pyramid attention for deep convolutional neural networks | |
| Shamsolmoali et al. | High-dimensional multimedia classification using deep CNN and extended residual units | |
| Ma et al. | Error correcting input and output hashing | |
| Tong et al. | A deep discriminative and robust nonnegative matrix factorization network method with soft label constraint | |
| Chakarverti et al. | Classification technique for heart disease prediction in data mining | |
| CN112685635A (en) | Item recommendation method, device, server and storage medium based on classification label | |
| Majumdar et al. | Cold-start, warm-start and everything in between: An autoencoder based approach to recommendation | |
| CN111310462A (en) | User attribute determination method, device, equipment and storage medium | |
| Liu et al. | Adaptive Group Sparse Multi-task Learning via Trace Lasso. | |
| Yang et al. | Unsupervised feature selection based on reconstruction error minimization | |
| CN110389932A (en) | Electric power automatic document classifying method and device | |
| Roy et al. | Design of movie recommendation system by means of collaborative filtering | |
| CN113486240A (en) | Position recommendation method based on SWPEM routing algorithm | |
| CN112734519B (en) | Commodity recommendation method based on convolution self-encoder network | |
| Cao et al. | Fuzzy emotional semantic analysis and automated annotation of scene images | |
| WO2017168601A1 (en) | Similar image search method and system | |
| Choi et al. | An efficient incremental face annotation for large scale web services |
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 |