CN106709758A - Intelligent recommendation and distribution solution for service class commodities in electronic commerce products - Google Patents
Intelligent recommendation and distribution solution for service class commodities in electronic commerce products Download PDFInfo
- Publication number
- CN106709758A CN106709758A CN201611179838.7A CN201611179838A CN106709758A CN 106709758 A CN106709758 A CN 106709758A CN 201611179838 A CN201611179838 A CN 201611179838A CN 106709758 A CN106709758 A CN 106709758A
- Authority
- CN
- China
- Prior art keywords
- user
- attendant
- service
- implicit features
- matrix
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0202—Market predictions or forecasting for commercial activities
Landscapes
- Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Engineering & Computer Science (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Finance (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- Marketing (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an intelligent recommendation and distribution solution for service class commodities in electronic commerce products, which is characterized in that analysis is performed on order data and service worker data through an OIT intelligent algorithm according to factors such as on-duty conditions, skills and identities of service workers, the distance between a site where the service workers are located and an order service address and idle conditions by combining information such as attributive classification of the service class commodities, the member type, whether a service worker is specified or not, whether multiple service workers are required or not and whether an order is not received overtime or not, thereby automatically completing distribution in a one-to-one or one-to-many mode. According to the method, services are provided for customers more accurately and more quickly on the one hand; and on the other hand, the working time of the service workers can be coordinated and distributed more reasonably, more effectively and more intelligently, and the service capacity and the service level of the whole service team are improved.
Description
Technical field
The invention provides intelligent recommendation and distribution solution that class commodity are serviced in a kind of electric business product, it is related to commodity
Core technology and the algorithm such as classification, attendant's division of labor, LBS positioning and calculating, intelligent order matching, efficiently carry out order point
With scheduling.
Background technology
With the development of Internet technology, the magnanimity information epoch have arrived.From another aspect, nowadays, O2O makees
It is a kind of pattern important in ecommerce, the every aspect of our lives has been permeated already.Purchased by group from earliest primary application
Business all approve the ecommerce of this type with the business such as food and drink, increasing consumer in recent years most active calling a taxi,
Select the service admired again to enjoyment service under line on line.For O2O markets, service for life class O2O markets need higher
INF, localization degree and network of personal connections, are the fields relatively weak for solid financial strength giants, are vast medium and small foundation
Person provides has relatively low entry threshold, investment value higher and gold development space.
The content of the invention
Class commodity are serviced it is an object of the invention to be directed in electric business product, is that medium and small pioneering enterprise or individual services are carried
Donor provides a kind of O2O intelligent recommendations and distribution solution.
The purpose of the present invention is achieved through the following technical solutions:The intelligence of class commodity is serviced in a kind of electric business product
Recommend and distribution solution, comprise the following steps:
(1) data cleansing.User, the attribute information of attendant and the user of history are extracted from database for clothes
The evaluation information of business personnel;
(2) feature extraction.The user of history evaluation information architecture rating matrix according to to(for) attendant;To user and thing
Product build the implicit features matrix of user and attendant respectively.Wherein, the implicit features matrix representative of user user for
The preference of service, and the implicit features matrix of attendant represents degree of the attendant comprising these preferences.User concealed spy
Levy vectorial UiWith attendant's implicit features vector VjForm it is as follows:
Ui=α Pi+εi
Vj=β Qj+εj
Wherein, PiRepresent the attribute feature vector of user i, QjThe attribute feature vector of attendant j is represented, α represents PiIt is right
UiInfluence degree, β represents QjTo VjInfluence degree;εi、εjIt is variance.
(3) model training.Random initializtion implicit features matrix;The characteristic of attribute feature vector is considered based on CAT algorithms,
User and the corresponding implicit features matrix U of attendant, V are obtained using stochastic gradient descent algorithm (SGD);
(4) integrated study.Using Bagging integrated learning approachs, m basis is trained by way of sample resampling
Model;The rating matrix of these model predictions is carried out averagely, to obtain final rating matrix;
(5) recommendation process.According to m basic model, the implicit spy of user to be recommended and all attendants is calculated respectively
The inner product of vector is levied, prediction scoring of the m groups user for all potential attendants is obtained, user is obtained and is taken for each
The m average value of prediction scoring of business personnel, is ranked up according to these scorings, filters out the clothes for being unsatisfactory for user's querying condition
Business personnel, choose K scoring highest attendant and recommend user, for user's selection.
Further, model online updating step is also included after the step (5).When a new scoring records generation
When, user and the corresponding implicit features vector of attendant are updated, by stochastic gradient algorithm's m basic model of random selection
One be updated.
Further, user's querying condition includes the distance of skill set requirements, service time, user and attendant.
The beneficial effects of the invention are as follows:
(1) with the rise of O2O business models, increasing user's selection is consumed on the net.The present invention is directed to
This Special Category of service class product, research and analysis have been carried out for its particularity in electric business product;Current service class is produced
Product recommend method, typically simply by calculating the service time of user input and using LBS to calculate relative distance to push away
Attendant is recommended, although this mode is simple, many additional informations in association and the system between user is have ignored.
Because similar user may be to similar attendant more preference, the present invention uses CAT algorithms, by user and attendant
Attributive character be added in collaborative filtering, introduce user concealed eigenmatrix and attendant's implicit features matrix
Concept;
(2) by implicit features matrix, we can soon obtain prediction scoring of the user for all attendants,
Here it is contemplated that to the noise problem of production environment, using Bagging Ensemble Learning Algorithms, by safeguarding m model so that
Predicting the outcome more has robustness;
(3) because the user's number and attendant's number in electric business platform are numerous, when a new evaluation is generated, I
By on-line learning algorithm, randomly update 1 model using stochastic gradient descent method, so both reduced the meter of commending system
Burden is calculated, the robustness of bagging integrated studies is also maintained.
Brief description of the drawings
Fig. 1 is user-attendant contacts figure;
Fig. 2 is to use CAT algorithms and Bagging integrated approach training patterns;
Fig. 3 is recommended flowsheet and model online updating.
Specific embodiment
The present invention is described in further detail with specific embodiment below in conjunction with the accompanying drawings.
One typical service class commodity platform should include following principal character:
(1) user, comprising attributes such as age, sex, location, member's types;
(2) attendant, comprising attributes such as age, sexes;
(3) order, comprising attributes such as service time, skill set requirements;
(4) evaluate, evaluation of the user for attendant.
Situation on duty according to attendant of the invention, technical ability, identity, attendant location to order placement service address
The factors such as distance, idle condition, with reference to the attributive classification of service class commodity, member's type, the information such as order specified services time,
Order data and service demographic data are analyzed by OIT intelligent algorithms, recommend Top K attendants to supply user to user
Selection.Specifically include following steps:
(1) data cleansing.User, the attribute information of attendant and the user of history are extracted from database for clothes
The evaluation information of business personnel;
(2) feature extraction.The user of history evaluation information architecture rating matrix according to to(for) attendant;To user and thing
Product build the implicit features matrix of user and attendant respectively.Wherein, the implicit features matrix representative of user user for
The preference of service, and the implicit features matrix of attendant represents degree of the attendant comprising these preferences.User concealed spy
Levy vectorial UiWith attendant's implicit features vector VjForm it is as follows:
Ui=α Pi+εi
Vj=β Qj+εj
Wherein, PiRepresent the attribute feature vector of user i, QjThe attribute feature vector of attendant j is represented, α represents PiIt is right
UiInfluence degree, β represents QjTo VjInfluence degree;εi、εjIt is variance.
(3) model training.Object function form is as follows:
RijRepresent true scorings of the user i for attendant j;
λ represents regularization coefficient (Controlling model complexity prevents the generation of over-fitting).
Random initializtion implicit features matrix;
The characteristic of attribute feature vector is considered based on CAT algorithms, all of scoring record is traveled through, stochastic gradient descent is used
Algorithm (SGD) obtains user and the corresponding implicit features matrix U of attendant, V;
Scored for wall scroll and recorded, the more new formula of SGD is as follows:
Wherein, η represents learning rate.
(4) integrated study.Because the data set noise ratio in production environment is higher, in order to reduce noise for model prediction
The influence of accuracy, we use Bagging integrated learning approachs, and m basic mould is trained by way of sample resampling
Type.
(5) recommendation process.According to m basic model, the implicit spy of user to be recommended and all attendants is calculated respectively
The inner product of vector is levied, prediction scoring of the m groups user for all potential attendants is obtained, user is obtained and is taken for each
The m average value of prediction scoring of business personnel, is ranked up according to these scorings, filters out the clothes for being unsatisfactory for user's querying condition
Business personnel, choose K scoring highest attendant and recommend user, for user's selection.
Further, model online updating step is also included after the step (5).When a new scoring records generation
When, user and the corresponding implicit features vector of attendant are updated, by stochastic gradient algorithm's m basic model of random selection
One be updated.
Further, user's querying condition includes the distance of skill set requirements, service time, user and attendant.
We introduce the specific implementation step of this invention with reference to an example for online business of keeping a public place clean below.
As shown in figure 1, the online business packet two kinds of roles containing user and attendant that keep a public place clean, and user's inclusive
Not, the build-in attribute such as age, area, member's type, attendant also has the attributes such as age, sex, affiliated unit.Work as user
When having demand for services, the restrictive conditions such as skill set requirements, service time can be also given.The following is the present invention in this special scenes
With:
(1) history evaluation information of the user to the auntie that keeps a public place clean is extracted from MySQL database, rating matrix X, scoring is formed
Scope (1-5), " " represent user not yet auntie is scored, XijScorings of the user i to the auntie j that keeps a public place clean is represented, such as the institute of table 1
Show.
The user of table 1-auntie's rating matrix of keeping a public place clean
(2) attribute information of user and the auntie that keeps a public place clean is extracted, as Table 2,3, wherein this for the age successional
Feature, we can be by its discretization (calculating an age bracket in such as every 10 years), and for this attribute of affiliated unit, we can make
Represented with binary-coding (binarycoding).
The customer attribute information of table 2
Table 3 is kept a public place clean auntie's attribute information
(3) the implicit features matrix to user and the auntie that keeps a public place clean is initialized, and adds user and the Ah that keeps a public place clean wherein
The influence of the attribute feature vector of one's mother's sister:
Ui=α Pi+εi
Vj=β Qj+εj
(4) based on CAT algorithms, optimize object function mentioned above, user is obtained using stochastic gradient descent calligraphy learning
Implicit features matrix U and the implicit features matrix V of auntie of keeping a public place clean.
(5) Bagging integrated studies.Sample resampling is carried out for the scoring record of the auntie that keeps a public place clean to user, m is obtained
Different training sets, trains m basic model.
(6) after the completion of model training, for user i we by calculating m model
Pi=Ui*VT
Prediction scoring of the user for all aunties that keep a public place clean is obtained, then the predicted value to this m model is averaged, and obtains
Scored to last prediction.Then we are filtered according to conditions such as skill set requirements, the service times of user input, further according to
The distance of current keep a public place clean auntie and user that LBS is calculated, filters out away from auntie too far away.Finally we are according to pre- test and appraisal
Divide and be ranked up from high to low, choose the Top K auntie that keeps a public place clean and recommend user;
(6) when a new scoring records generation, the implicit features square of user is updated using stochastic gradient descent method
Battle array, is updated by the basic model randomly choosed in bagging integrated approaches, so both reduces model modification
Amount of calculation, while also improving the robustness of model, reduces susceptibility of the integrated model for noise.The implementation of whole invention
Step is as shown in Figure 2,3.
Claims (3)
1. the intelligent recommendation and distribution solution of class commodity are serviced in a kind of electric business product, it is characterised in that including following step
Suddenly:
(1) data cleansing.User, the attribute information of attendant and the user of history are extracted from database for service people
The evaluation information of member;
(2) feature extraction.The user of history evaluation information architecture rating matrix according to to(for) attendant;To user and article point
Not Gou Jian user and attendant implicit features matrix.Wherein, the implicit features matrix representative of user user for service
Preference, and the implicit features matrix of attendant represents degree of the attendant comprising these preferences.User concealed feature to
Amount UiWith attendant's implicit features vector VjForm it is as follows:
Ui=α Pi+εi
Vj=β Qj+εj
Wherein, PiRepresent the attribute feature vector of user i, QjThe attribute feature vector of attendant j is represented, α represents PiTo Ui's
Influence degree, β represents QjTo VjInfluence degree;εi、εjIt is variance.
(3) model training.Random initializtion implicit features matrix;The characteristic of attribute feature vector is considered based on CAT algorithms, is used
Stochastic gradient descent algorithm (SGD) obtains user and the corresponding implicit features matrix U of attendant, V;
(4) integrated study.Using Bagging integrated learning approachs, m basic mould is trained by way of sample resampling
Type;The rating matrix of these model predictions is carried out averagely, to obtain final rating matrix;
(5) recommendation process.According to m basic model, calculate respectively the implicit features of user to be recommended and all attendants to
The inner product of amount, obtains prediction scoring of the m groups user for all potential attendants, obtains user for each service people
The m average value of prediction scoring of member, is ranked up according to these scorings, filters out the service people for being unsatisfactory for user's querying condition
Member, chooses K scoring highest attendant and recommends user, for user's selection.
2. the intelligent recommendation and distribution solution of class commodity are serviced in a kind of electric business product according to claim 1, its
It is characterised by, model online updating step is also included after the step (5).When a new scoring records generation, update
User and the corresponding implicit features vector of attendant, in m basic model is randomly choosed by stochastic gradient algorithm
It is updated.
3. the intelligent recommendation and distribution solution of class commodity are serviced in a kind of electric business product according to claim 1, its
It is characterised by, user's querying condition includes the distance of skill set requirements, service time, user and attendant.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611179838.7A CN106709758A (en) | 2016-12-19 | 2016-12-19 | Intelligent recommendation and distribution solution for service class commodities in electronic commerce products |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611179838.7A CN106709758A (en) | 2016-12-19 | 2016-12-19 | Intelligent recommendation and distribution solution for service class commodities in electronic commerce products |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106709758A true CN106709758A (en) | 2017-05-24 |
Family
ID=58938371
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611179838.7A Pending CN106709758A (en) | 2016-12-19 | 2016-12-19 | Intelligent recommendation and distribution solution for service class commodities in electronic commerce products |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106709758A (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107748984A (en) * | 2017-10-31 | 2018-03-02 | 广州为想互联网科技有限公司 | Automatic job distribution method and device |
CN107977825A (en) * | 2017-11-03 | 2018-05-01 | 阿里巴巴集团控股有限公司 | A kind of method and device for distributing Service events |
CN107977794A (en) * | 2017-12-14 | 2018-05-01 | 方物语(深圳)科技文化有限公司 | Data processing method, device, computer equipment and the storage medium of industrial products |
CN109191236A (en) * | 2018-06-15 | 2019-01-11 | 长沙市到家悠享家政服务有限公司 | Order generation method, device and electronic equipment |
CN109886557A (en) * | 2019-01-25 | 2019-06-14 | 深圳微品致远信息科技有限公司 | A kind of order distribution method and system based on big data and gridding |
CN111080098A (en) * | 2019-12-04 | 2020-04-28 | 中国太平洋保险(集团)股份有限公司 | O2O intelligent matching algorithm and device |
CN111914165A (en) * | 2020-06-29 | 2020-11-10 | 长沙市到家悠享网络科技有限公司 | Target object recommendation method, device, equipment and storage medium |
CN112348623A (en) * | 2020-09-14 | 2021-02-09 | 长沙市到家悠享网络科技有限公司 | Information processing method, device and storage medium |
CN113379457A (en) * | 2021-06-04 | 2021-09-10 | 浙江杭州余杭农村商业银行股份有限公司 | Intelligent marketing method oriented to financial field |
CN115640465A (en) * | 2022-12-26 | 2023-01-24 | 北京璐珠科技有限公司 | Cross-region and cross-merchant resource sharing method and system |
CN117391405A (en) * | 2023-12-11 | 2024-01-12 | 汇丰金融科技服务(上海)有限责任公司 | Method, system and electronic device for intelligent matching of clients and business personnel |
-
2016
- 2016-12-19 CN CN201611179838.7A patent/CN106709758A/en active Pending
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107748984A (en) * | 2017-10-31 | 2018-03-02 | 广州为想互联网科技有限公司 | Automatic job distribution method and device |
CN107977825A (en) * | 2017-11-03 | 2018-05-01 | 阿里巴巴集团控股有限公司 | A kind of method and device for distributing Service events |
CN107977794A (en) * | 2017-12-14 | 2018-05-01 | 方物语(深圳)科技文化有限公司 | Data processing method, device, computer equipment and the storage medium of industrial products |
CN109191236A (en) * | 2018-06-15 | 2019-01-11 | 长沙市到家悠享家政服务有限公司 | Order generation method, device and electronic equipment |
CN109886557A (en) * | 2019-01-25 | 2019-06-14 | 深圳微品致远信息科技有限公司 | A kind of order distribution method and system based on big data and gridding |
CN111080098A (en) * | 2019-12-04 | 2020-04-28 | 中国太平洋保险(集团)股份有限公司 | O2O intelligent matching algorithm and device |
CN111914165A (en) * | 2020-06-29 | 2020-11-10 | 长沙市到家悠享网络科技有限公司 | Target object recommendation method, device, equipment and storage medium |
CN111914165B (en) * | 2020-06-29 | 2023-10-20 | 长沙市到家悠享网络科技有限公司 | Target object recommendation method, device, equipment and storage medium |
CN112348623A (en) * | 2020-09-14 | 2021-02-09 | 长沙市到家悠享网络科技有限公司 | Information processing method, device and storage medium |
CN113379457A (en) * | 2021-06-04 | 2021-09-10 | 浙江杭州余杭农村商业银行股份有限公司 | Intelligent marketing method oriented to financial field |
CN115640465A (en) * | 2022-12-26 | 2023-01-24 | 北京璐珠科技有限公司 | Cross-region and cross-merchant resource sharing method and system |
CN117391405A (en) * | 2023-12-11 | 2024-01-12 | 汇丰金融科技服务(上海)有限责任公司 | Method, system and electronic device for intelligent matching of clients and business personnel |
CN117391405B (en) * | 2023-12-11 | 2024-03-15 | 汇丰金融科技服务(上海)有限责任公司 | Method, system and electronic device for intelligent matching of clients and business personnel |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106709758A (en) | Intelligent recommendation and distribution solution for service class commodities in electronic commerce products | |
Zhang et al. | An examination of social influence on shopper behavior using video tracking data | |
Ivanov et al. | Hotel revenue management–a critical literature review | |
Guo et al. | Applications of artificial intelligence in the apparel industry: a review | |
Cil | Consumption universes based supermarket layout through association rule mining and multidimensional scaling | |
CN105225129B (en) | Mobile O2O recommendation method and system thereof | |
CN106651524A (en) | Method for intelligently generating recommended menu | |
CN109615184A (en) | The method and system of shops, retailer automatic cargo allocation, the goods that replenishes, adjusts | |
Bala et al. | Association rule for purchase dependence in multi-item inventory | |
CN102282551A (en) | Automated decision support for pricing entertainment tickets | |
Behera et al. | Impact of store location and layout on consumer purchase behavior in organized retail | |
JP2019028899A (en) | Sales promotion system, machine learning device, machine learning data providing device, and program | |
US11587028B2 (en) | Computer system for optimizing garment inventory of retailer based on shapes of users | |
CN116645141B (en) | Multi-dimensional feature-fused chain store site selection recommendation method | |
Hernández-Blanco et al. | Natural capital and ecosystem services | |
JP2017220155A (en) | Customer management in electronic commerce site | |
CN113076484A (en) | Product recommendation method, device, equipment and storage medium based on deep learning | |
Kesavan et al. | Increasing sales by managing congestion in self-service environments: Evidence from a field experiment | |
Yang et al. | Forecast UPC-level FMCG demand, Part II: Hierarchical reconciliation | |
Fitzpatrick et al. | Agent-based modeling of ecological niche theory and assortative drinking | |
US20210287275A1 (en) | Value based commodity selection | |
Kaur et al. | Market basket analysis of sports store using association rules | |
CN113641904A (en) | Method and device for recommending target customers for stores under Internet marriage line | |
WO2021065289A1 (en) | Store assistance system, store assistance method, and program | |
Rezaeinia et al. | An integrated AHP-RFM method to banking customer segmentation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170524 |