CN108665323B - Integration method for financial product recommendation system - Google Patents
Integration method for financial product recommendation system Download PDFInfo
- Publication number
- CN108665323B CN108665323B CN201810484714.2A CN201810484714A CN108665323B CN 108665323 B CN108665323 B CN 108665323B CN 201810484714 A CN201810484714 A CN 201810484714A CN 108665323 B CN108665323 B CN 108665323B
- Authority
- CN
- China
- Prior art keywords
- user
- item
- preference
- users
- recommendation
- 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.)
- Active
Links
Images
Classifications
-
- 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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0255—Targeted advertisements based on user history
-
- 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
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
-
- 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
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/06—Asset management; Financial planning or analysis
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Human Resources & Organizations (AREA)
- Operations Research (AREA)
- Technology Law (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses an integration method for a financial product recommendation system, which can fill sparse data based on a collaborative filtering algorithm of data smoothing and reduce the problem of data sparsity. The recommendation algorithm based on the demographics does not need historical data and does not depend on the attribute of an article, so that the cold start problem of a user can be solved; the two algorithms are integrated with a recommendation algorithm which has good performance and is based on item clustering and matrix decomposition, so that the use scene of the recommendation algorithm is enlarged, and the adaptability of the recommendation algorithm is improved. The integration method can effectively reduce the sparsity of data, solve the problem of cold start and improve the recommendation performance of each user.
Description
Technical Field
The invention belongs to the technical field of online product recommendation, and particularly relates to an integration method for a financial product recommendation system.
Background
The traditional recommendation method mostly calculates interest preference and resource similarity of users according to user score data, has low recommendation quality for sparse data and new users, and cannot maximally mine information carried by implicit data.
In recent years, aiming at the problems brought by data sparsity, in order to improve the recommendation effect, students introduce algorithms such as principal component analysis, cluster analysis, singular value decomposition and the like into a traditional collaborative filtering recommendation algorithm, and reduce the range of searching nearest neighbors by a target user through dimension reduction, so that the recommendation precision and real-time performance are obviously improved, but the cold start problem still exists.
The main structure and principle of the prior art are as follows:
1. a user-item attribute preference model is generated. The preference model of the user for the project attributes is the basis for user clustering and similarity calculation, and a preference weight matrix of the user for all attributes appearing in the project is established by analyzing a user-project scoring matrix and a project-attribute matrix.
2. And (5) clustering users. Clustering users by adopting a mixed clustering model combining SOM and K-means clustering:
2.1 taking the obtained user-item attribute preference matrix as input data of clustering, carrying out coarse clustering on input training for a few times through SOM, and outputting ClusterSOMWeight omega of neuronSOMThe number of clustering clusters K;
2.2 will ωSOMAs the original centroid OoriginalFor each cluster with element not 0, find and OoriginalThe closest element is taken as the final centroid O of the clusterSOM;
2.3 at K, OSOMAnd further clustering the users as the clustering cluster number and the initial clustering center of mass of the K-means cluster, and outputting a clustering result ClusterResult of the users.
3. User similarity calculation and nearest neighbor query. Calculating target user U by cosineiAnd the cluster cindexThe similarity of other users in the system is calculated to obtain a nearest neighbor set MKnear。
4. And (4) score prediction. Finding target user UiFor target item IijM of nearest neighbor usersKnearThen, through the set MKnearFor the target item IijWeighted average of scores to describe target user UiFor target item IijThe score of (1). The score prediction formula is shown as formula (1):
5. a recommendation is generated. Repeating the step (3) and the step (4), and predicting the target user UiSelecting N items with highest predicted scores to recommend to a target user U for the scores of all the unscored itemsi。
There are many recommendation algorithms at present, but there is no one algorithm that always outperforms the others in any context or under any data. The existing recommendation systems are mostly single methods, have self limitations and cannot be flexibly applied to various scenes. Algorithms that perform well in recommendation cannot effectively address data sparsity and cold start issues.
Disclosure of Invention
The invention provides an integration method for a financial product recommendation system, which reduces the influence caused by data sparsity and solves the cold start problem of the recommendation system. The adaptability of the recommendation algorithm is improved, and the application scene of the recommendation algorithm is expanded.
The collaborative filtering algorithm based on data smoothing can fill sparse data, and the problem of data sparsity is reduced. The demographic-based recommendation algorithm does not require historical data nor rely on the attributes of the items, and can solve the cold start problem of the user. The two algorithms are integrated with a recommendation algorithm which has good performance and is based on item clustering and matrix decomposition, so that the use scene of the recommendation algorithm is enlarged, and the adaptability of the recommendation algorithm is improved. The integration method can effectively reduce the sparsity of data, solve the problem of cold start and improve the recommendation performance of each user.
Drawings
FIG. 1 is a flow chart of an integration method for a financial product recommendation system according to the present invention.
Detailed Description
As shown in fig. 1, the present invention provides an integrated method for a financial product recommendation system, comprising the steps of:
the input data is: user characteristic information, a project-attribute matrix and a user-project scoring matrix; the output data is: a product recommendation model and a user recommendation result.
The method comprises the following steps: a demographic based recommendation algorithm.
The traditional recommendation algorithm based on demographics is improved, and different user attributes are endowed with different weights. The invention selects 4 characteristics of age, sex, occupation and hobby as the considered range, and preprocesses each attribute information into the form of digital representation. And calculating the similarity among the users to obtain the user preference.
1. Age attribute, the present invention is in 5 years increments, such as 32 years old may be scored as 7 and 56 years old as 12. And (3) calculating the similarity of the ages between the user s and the user t by using the Euclidean distance, wherein the similarity is shown as a formula (2).
Wherein A (s, t) representsSimilarity of user s and user t in age, xs、xtAge segmentation values for users s and t, respectively.
2. The distinction is a symmetric binary property, i.e. both states are equally important. The male is marked as 1, and the female is marked as 0, so that a list of two rows and two columns can be obtained according to the value of the gender of the user, as shown in table 1. Here, S (S, t) is used to represent the similarity between the user S and the user t in terms of gender, as shown in equation (3), where a is an attribute value in the case where both the object S and t take 1, b is an attribute value in the case where the object t takes 1 and the object S takes 0, c is an attribute value in the case where the object t takes 0 and the object S takes 1, and d is an attribute value in the case where both the object S and t take 0.
Table 1 binary attribute values
3. Occupation and hobby belong to the attribute of the label type, namely, the attribute is described by certain fields, and for the attribute, the similarity between the users s and t is calculated by using an equation (4). In the formula (4), k represents the number of tags having the same attribute and shared by the user s and t, and n represents the number of tags selectable for the attribute.
4. And calculating the similarity of the users. For user U and k neighbor sets UkEquation (5) calculates the information similarity of users and saves Top-N user sets U similar to the target user Uk. Under the user demographic feature method, the more features, the more accurate the user interest can be predicted.
Calculating user preferences
After the similarity between the users is obtained, K favorite articles of the users with the most similar interests to the users are recommended to the users, and the preference of the user u for the article i is calculated by adopting the following formula:
wherein S (u, K) includes K users closest to user u, N (i) is a set of users having past behavior on item i, wuvIs the interest similarity of user u and user v, rviIndicating the preference of user v for item i.
Deriving a prediction preference matrix
n is the number of items and m is the number of users.
Step two: recommendation algorithm based on item clustering and matrix decomposition
1. Calculating similarity between items
The distance between the items is calculated by using the manhattan distance.
ruiIndicating the user u's preference for item i. dijRepresenting the distance between item i and item j.
The similarity between the item i and the item j is expressed by equation (9).
ciIndicates the popularity of item i, cjIndicating the popularity of item j. Popularity of articlesThe degree is the number of people clicking on the item.
Then classifying the articles to obtain different clustering centers { c1,c2,Λ,ckAnd k is the number of clusters.
2. An item vector is constructed.
Based on K cluster centers, K is set to 200, and the item vector is defined as
Wherein the content of the first and second substances,
normalizing the article vector:
finally, the vector for item i is:
pi=(pi1,pi2,Λ,pik,Λ,piK) (13)
wherein the content of the first and second substances,
3. a prediction preference matrix is calculated. Based on the commodity vector and Singular Value Decomposition (SVD), a prediction preference matrix can be derived:
n is the number of items, m is the number of users,represents the prediction preference, defined as:
where allMean is the average of the preferences, buRepresenting the deviation between the user and the allMean, biRepresenting the deviation between the item and the allMean, quIs a vector of users u, initialized with random values.
Step three: collaborative filtering algorithms based on data smoothing.
1. And calculating the similarity of the users. The similarity is calculated using the pearson correlation coefficient.
The similarity between user u and user u' is:
Ru(t) represents a preference of user u for item t,representing the average preference of user u for all items, Ru’(t) represents a preference of user u' for item t,representing the average preference of user u 'for all items, and t is the item that user u and user u' both clicked on.
2. Based on the previous step, the data set that the user has not clicked on is smoothed. The user's preferences are expressed as:
wherein r isuiIs calculated by a function, and is obtained by calculating,it is for i that user u has not clicked on that is derived from the smoothing.
For user u, the cluster to which u belongs is represented asIn consideration of individual differences, the calculation is performed by equation (19)
wherein, Cu(i)∈CuIs represented in cluster CuUser set of item i clicked on, | Cu(i) I is represented in cluster CuThe number of users who clicked on item i.
The prediction preference may be derived by calculating a weighted sum:
is the preference of user u for item i,is the average preference of the item i,is the average preference of item j, wujIs the weight between u and j, and sim (i, j) is the similarity of i and j.
3. A prediction preference matrix is obtained.
n is the number of items and m is the number of users.
Step four: integrating algorithms
According to the first step, the second step and the third step, the prediction preference of the user for each product is obtained, and algorithm integration is carried out based on the preferences.
1. Linear weighted fusion method
Summarizing the results of a single model, then giving different weights according to different algorithms, and weighting the results of a plurality of recommendation algorithms to obtain the results:
2. Method of cross fusion
And inserting results of different recommendation models in the recommendation result to ensure the diversity of the results.
rec (u) denotes an item recommended to the user u, reck(u) represents the item recommended by algorithm k to user u.
3. Waterfall fusion method
The waterfall type fusion method adopts a method of connecting a plurality of models in series. Each recommendation algorithm is regarded as a filter and is performed by a method of linking filters with different granularities back and forth, in the method, the result filtered by the former recommendation method is input as a candidate set of the latter recommendation method, the candidate result is progressively selected in the process, and finally a recommendation result set with few high quality is obtained.
Claims (2)
1. An integration method for a financial product recommendation system, comprising the steps of:
the method comprises the following steps: a demographic-based recommendation algorithm;
selecting 4 characteristics of age, gender, occupation and hobby, preprocessing each attribute information into a digital representation form, calculating the similarity between users to obtain user preference and a prediction preference matrix;
step two: recommendation algorithm based on item clustering and matrix decomposition
Step 2.1, calculating similarity between articles
Calculating a distance between the articles by using the manhattan distance;
wherein r isuiIndicates the user u's preference for item i, dijRepresenting the distance between item i and item j,
the similarity between the item i and the item j is expressed by formula (9);
wherein, ciIndicates the popularity of item i, cjRepresents the popularity of item j; the popularity of an item is clicking on itThe number of people who have articles, then the articles are classified to obtain different clustering centers { c1,c2,…,ckK is the number of clusters;
step 2.2, construct the item vector
Based on K cluster centers, K is set to 200, and the item vector is defined as
Wherein the content of the first and second substances,
normalizing the article vector:
finally, the vector for item i is:
pi=(pi1,pi2,…,pik,…,piK) (13)
wherein the content of the first and second substances,
step 2.3, calculating a prediction preference matrix
Based on the item vector and Singular Value Decomposition (SVD), a prediction preference matrix is obtained:
where n is the number of items, m is the number of users,represents the prediction preference, defined as:
where allMean is the average of the preferences, buRepresenting the deviation between the user and the allMean, biRepresenting the deviation between the item and the allMean, quIs the vector of user u, initialized by random values;
step three: collaborative filtering algorithm based on data smoothing
Step 3.1, calculating the similarity of users
And calculating the similarity by adopting a Pearson correlation coefficient, wherein the similarity between the user u and the user u' is as follows:
wherein R isu(t) represents a preference of user u for item t,representing the average preference of user u for all items, Ru’(t) represents a preference of user u' for item t,representing the average preference of the user u 'for all items, and t is the item clicked by both the user u and the user u';
Step 3.2, smoothing the data set which is not clicked by the user based on the previous step
The user's preferences are expressed as:
wherein r isuiIs calculated by a function, and is obtained by calculating,is derived from the smoothing for item i that user u has not clicked,
for user u, the cluster to which u belongs is represented asIn consideration of individual differences, the calculation is performed by equation (19)
wherein, Cu(i)∈CuIs represented in cluster CuUser set of item i clicked on, | Cu(i) I is represented in cluster CuThe number of users who clicked on item i,
the prediction preference is obtained by calculating a weighted sum:
wherein the content of the first and second substances,is the preference of user u for item i,is the average preference of the item i,is the average preference of item j, wujIs the weight between u and j, sim (i, j) is the similarity of i and j;
step 3.3, obtaining a prediction preference matrix
n is the number of items, m is the number of users;
step four: integrating algorithms
According to the first step, the second step and the third step, the prediction preference of the user for each product is obtained, and algorithm integration is carried out based on the preferences;
in the first step, the calculation process of the user preference is as follows:
after the similarity between the users is obtained, K favorite articles of the users with the most similar interests to the users are recommended to the users, and the preference of the user u for the article i is calculated by adopting the following formula:
wherein S (u, K) includes K users closest to user u, N (i) is a set of users having past behavior on item i, wuvIs the interest similarity of user u and user v, rviRepresenting a preference of user v for item i;
the calculation process of the prediction preference matrix is as follows:
where n is the number of items and m is the number of users.
2. The integration method for a financial product recommendation system according to claim 1, wherein step four is integrated by:
1) linear weighted fusion method
Summarizing the results of a single model, then giving different weights according to different algorithms, and weighting the results of a plurality of recommendation algorithms to obtain the results:
wherein the content of the first and second substances,is the predicted final preference, wkIs the weight corresponding to the k algorithm;
2) method of cross fusion
The results of different recommendation models are interspersed in the recommendation result to ensure the diversity of the results,
wherein rec (u) denotes an item recommended to the user u, reck(u) represents the recommended items for user u by algorithm k;
3) waterfall type fusion method
The waterfall type fusion method adopts a method of connecting a plurality of models in series, each recommendation algorithm is regarded as a filter, and the method is carried out by connecting filters with different granularities back and forth.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810484714.2A CN108665323B (en) | 2018-05-20 | 2018-05-20 | Integration method for financial product recommendation system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810484714.2A CN108665323B (en) | 2018-05-20 | 2018-05-20 | Integration method for financial product recommendation system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108665323A CN108665323A (en) | 2018-10-16 |
CN108665323B true CN108665323B (en) | 2021-01-05 |
Family
ID=63776873
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810484714.2A Active CN108665323B (en) | 2018-05-20 | 2018-05-20 | Integration method for financial product recommendation system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108665323B (en) |
Families Citing this family (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109544231A (en) * | 2018-11-21 | 2019-03-29 | 浙江工商大学 | Logistics distribution service personalization recommended method based on Logistics Information Platform |
CN109902235B (en) * | 2019-03-06 | 2023-07-07 | 太原理工大学 | User preference clustering collaborative filtering recommendation algorithm based on bat optimization |
CN109977315A (en) * | 2019-03-29 | 2019-07-05 | 厦门铠甲网络股份有限公司 | A kind of article recommended method, device, equipment and storage medium |
CN112241894A (en) * | 2019-07-16 | 2021-01-19 | 百度时代网络技术(北京)有限公司 | Content delivery method and device and terminal |
CN110930259B (en) * | 2019-11-15 | 2023-05-26 | 安徽海汇金融投资集团有限公司 | Credited recommendation method and system based on mixed strategy |
CN110930226A (en) * | 2019-11-26 | 2020-03-27 | 中国建设银行股份有限公司 | Financial product recommendation method and device, electronic equipment and storage medium |
CN111179037B (en) * | 2019-12-31 | 2023-07-25 | 中国银行股份有限公司 | Financial product recommendation method and device |
CN111652669A (en) * | 2020-04-15 | 2020-09-11 | 珠海小礼鱼科技有限公司 | Intelligent terminal shopping guide system and method |
CN111552852B (en) * | 2020-04-27 | 2021-09-28 | 北京交通大学 | Article recommendation method based on semi-discrete matrix decomposition |
CN111966907A (en) * | 2020-08-21 | 2020-11-20 | 贝壳技术有限公司 | User preference cold start method, device, medium and electronic equipment |
CN113379482B (en) * | 2021-05-28 | 2023-12-01 | 车智互联(北京)科技有限公司 | Article recommendation method, computing device and storage medium |
CN113393297A (en) * | 2021-06-17 | 2021-09-14 | 中国工商银行股份有限公司 | Financial product pushing method and device |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2011096255A (en) * | 2009-10-30 | 2011-05-12 | Nec (China) Co Ltd | Ranking oriented cooperative filtering recommendation method and device |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103246672B (en) * | 2012-02-09 | 2016-06-08 | 中国科学技术大学 | User is carried out method and the device of personalized recommendation |
CN102968506A (en) * | 2012-12-14 | 2013-03-13 | 北京理工大学 | Personalized collaborative filtering recommendation method based on extension characteristic vectors |
CN106997347A (en) * | 2016-01-22 | 2017-08-01 | 华为技术有限公司 | Information recommendation method and server |
CN106846106A (en) * | 2017-01-16 | 2017-06-13 | 南京师范大学 | A kind of figure builds the matrix decomposition recommendation method under framework |
CN106980648B (en) * | 2017-02-27 | 2021-07-27 | 南京邮电大学 | Personalized recommendation method based on probability matrix decomposition and combined with similarity |
-
2018
- 2018-05-20 CN CN201810484714.2A patent/CN108665323B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2011096255A (en) * | 2009-10-30 | 2011-05-12 | Nec (China) Co Ltd | Ranking oriented cooperative filtering recommendation method and device |
Also Published As
Publication number | Publication date |
---|---|
CN108665323A (en) | 2018-10-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108665323B (en) | Integration method for financial product recommendation system | |
CN108363804B (en) | Local model weighted fusion Top-N movie recommendation method based on user clustering | |
KR100997541B1 (en) | The method and apparatus for image recommendation based on user profile using feature based collaborative filtering to resolve new item recommendation | |
CN108537624B (en) | Deep learning-based travel service recommendation method | |
CN104834686B (en) | A kind of video recommendation method based on mixing semantic matrix | |
CN107944035B (en) | Image recommendation method integrating visual features and user scores | |
CN107894998B (en) | Video recommendation method and device | |
CN113158023B (en) | Public digital life accurate classification service method based on mixed recommendation algorithm | |
CN107220365A (en) | Accurate commending system and method based on collaborative filtering and correlation rule parallel processing | |
CN108897791B (en) | Image retrieval method based on depth convolution characteristics and semantic similarity measurement | |
CN115391576A (en) | Cluster-based federal graph learning method and system | |
CN106874355A (en) | The collaborative filtering method of social networks and user's similarity is incorporated simultaneously | |
CN110442618B (en) | Convolutional neural network review expert recommendation method fusing expert information association relation | |
CN112100512A (en) | Collaborative filtering recommendation method based on user clustering and project association analysis | |
CN109034953B (en) | Movie recommendation method | |
Liu et al. | Using collaborative filtering algorithms combined with Doc2Vec for movie recommendation | |
CN106339486A (en) | Image retrieval method based on incremental learning of large vocabulary tree | |
Anwar et al. | Mrec-crm: Movie recommendation based on collaborative filtering and rule mining approach | |
CN115712780A (en) | Information pushing method and device based on cloud computing and big data | |
Yang et al. | Recognizing cartoon image gestures for retrieval and interactive cartoon clip synthesis | |
CN116680363A (en) | Emotion analysis method based on multi-mode comment data | |
Chen et al. | A fuzzy matrix factor recommendation method with forgetting function and user features | |
CN108389113A (en) | A kind of collaborative filtering recommending method and system | |
CN107622071B (en) | Clothes image retrieval system and method under non-source-retrieval condition through indirect correlation feedback | |
CN115204967A (en) | Recommendation method integrating implicit feedback of long-term and short-term interest representation of user |
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 | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20210617 Address after: 200335 Room 301, building 1, 111 Fuquan Road, Changning District, Shanghai Patentee after: SHANGHAI DIGITAL CHINA INFORMATION TECHNOLOGY SERVICE Co.,Ltd. Address before: 100124 No. 100 Chaoyang District Ping Tian Park, Beijing Patentee before: Beijing University of Technology |