WO2018196424A1 - 推荐方法及装置 - Google Patents

推荐方法及装置 Download PDF

Info

Publication number
WO2018196424A1
WO2018196424A1 PCT/CN2017/118777 CN2017118777W WO2018196424A1 WO 2018196424 A1 WO2018196424 A1 WO 2018196424A1 CN 2017118777 W CN2017118777 W CN 2017118777W WO 2018196424 A1 WO2018196424 A1 WO 2018196424A1
Authority
WO
WIPO (PCT)
Prior art keywords
user
merchant
evaluation value
evaluation
recommended
Prior art date
Application number
PCT/CN2017/118777
Other languages
English (en)
French (fr)
Inventor
曾春
Original Assignee
北京小度信息科技有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 北京小度信息科技有限公司 filed Critical 北京小度信息科技有限公司
Publication of WO2018196424A1 publication Critical patent/WO2018196424A1/zh
Priority to US16/663,305 priority Critical patent/US20200126035A1/en

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q10/00Administration; Management
    • G06Q10/08Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
    • G06Q10/083Shipping
    • G06Q10/0834Choice of carriers
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90Details of database functions independent of the retrieved data types
    • G06F16/95Retrieval from the web
    • G06F16/953Querying, e.g. by the use of web search engines
    • G06F16/9535Search customisation based on user profiles and personalisation
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00Computing arrangements using knowledge-based models
    • G06N5/04Inference or reasoning models
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q30/00Commerce
    • G06Q30/02Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241Advertisements
    • G06Q30/0251Targeted advertisements
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00Machine learning

Definitions

  • the present disclosure relates to the field of Internet technologies, and in particular, to a recommendation method and apparatus.
  • personalized recommendation services are increasingly used in the Internet industry, such as finding nearby services on the map and searching for nearby restaurants in the takeaway application.
  • interesting content such as a service or a restaurant
  • behavior data such as user browsing information, historical orders, and product evaluations, determine user interests, and then recommend based on user interests.
  • the user interest-based recommendation methods commonly used in the related art mainly include a content-based recommendation method and a Collaborative Filtering Recommendation method.
  • the inventors of the present disclosure analyzed existing content-based recommendation methods and recommended methods of collaborative filtering.
  • the content-based recommendation method mainly uses natural language processing, artificial intelligence, probability statistics, and machine learning to perform content filtering, and strives to discover the user's interest and recommend products similar to the products that the user liked before.
  • Collaborative filtering recommendation method the core of which is mainly to analyze user interest, find similar users of the user in the user group, synthesize the evaluation of certain information by these similar users, form the user's preference for this information, and then proceed to the user recommend.
  • the performance of the above recommended method is already very good, but there are still problems that cannot be recommended or recommended.
  • the inventor of the present disclosure further conducted research and analysis, and the above recommended method considers factors such as geographical location, user interest, user behavior, etc., but ignores the influence of time on user behavior. For example, users want to eat differently for dinner and lunch, eat differently today and yesterday, and eat differently on weekends and in peacetime. It can be seen that in this scenario, if the existing recommendation method is used to recommend to the user that the things that were previously enjoyed or eaten are not suitable, for the diversity of the user's diet, the user should be recommended for something that has not been eaten recently. In addition, as time goes by, some things that have been eaten will become more fresh, so it is also necessary to recommend similar products that have been consumed before.
  • the inventor of the present disclosure provides a recommendation method, the core of which is: in the recommendation process, considering the influence of the time factor, based on the time influence characteristics of the recommended recommendation scene, correcting the initial evaluation value of the user, and then based on the correction
  • the post evaluation value is recommended to achieve the purpose of considering the time factor in the recommendation process, thereby improving the accuracy of the recommendation result.
  • the embodiment of the present disclosure provides a recommendation method, including:
  • the obtaining the modified evaluation value of the at least one user corresponding to each of the at least one candidate merchant includes:
  • the user-business modification evaluation matrix includes a modified evaluation value of at least one user corresponding to each of the plurality of merchants in the system, and the plurality of merchants include the at least one candidate merchant.
  • the user-business correction evaluation matrix is pre-built, including:
  • first user-merchant initial evaluation matrix includes product reviews of at least one user corresponding to each of the plurality of merchants Value sequence
  • the first user-merchant initial evaluation matrix is modified according to the time impact characteristic under the recommendation scenario to obtain the user-business modification evaluation matrix.
  • the first user-merchant initial evaluation matrix is modified according to a time-influence characteristic in the recommended scenario to obtain the user-business-correction evaluation matrix, including:
  • the time impact characteristic in the recommendation scenario is: a characteristic that a positive influence of a time factor on a recommendation process increases with time.
  • the product evaluation value sequence of the user U a corresponding to the merchant S k in the first user-merchant initial evaluation matrix The corresponding network behavior occurs in a time series of The time impact characteristic according to the recommended scenario and the time sequence of the network behavior occurrence
  • Calculating the product evaluation value sequence Corresponding time correction factor sequence include:
  • T now indicates the current time
  • T period represents a time period in the recommended scenario
  • the commodity evaluation value sequence is Time correction factor sequence corresponding thereto Multiplying to obtain a modified evaluation value of the user U a corresponding to the merchant S k in the user-merchant correction evaluation matrix, including:
  • v' ak represents a modified evaluation value of the user U a corresponding to the merchant S k .
  • the obtaining the modified evaluation value of the at least one user corresponding to each of the at least one candidate merchant includes:
  • the recommending the merchant to the to-be-recommended user based on the modified evaluation value of the at least one user corresponding to the at least one candidate merchant including:
  • the merchant-based collaborative filtering algorithm is used to recommend the merchant to the to-be-recommended user based on the revised evaluation value of the at least one user corresponding to the at least one candidate merchant.
  • the determining, according to the user to be recommended, the at least one candidate merchant including:
  • an embodiment of the present disclosure further provides a recommendation apparatus, including:
  • a determining module configured to determine at least one candidate merchant according to the user to be recommended
  • An obtaining module configured to acquire a modified evaluation value of at least one user corresponding to each of the at least one candidate merchant, wherein the modified evaluation value is obtained by modifying the initial evaluation value according to a time influence characteristic in the recommended scenario;
  • the recommendation module is configured to recommend the merchant to the to-be-recommended user based on the corrected evaluation value of the at least one user corresponding to each of the at least one candidate merchant.
  • the acquiring module is specifically configured to:
  • the user-business modification evaluation matrix includes a modified evaluation value of at least one user corresponding to each of the plurality of merchants in the system, and the plurality of merchants include the at least one candidate merchant.
  • the device further includes:
  • a building module configured to pre-build the user-business correction evaluation matrix
  • the building module includes:
  • a dimension switching sub-module configured to perform dimension switching on the first user-commodity evaluation matrix to obtain a first user-merchant initial evaluation matrix, where the first user-merchant initial evaluation matrix includes each of the plurality of merchants Corresponding at least one user's product evaluation value sequence;
  • the correction submodule is configured to modify the first user-merchant initial evaluation matrix according to the time impact characteristic in the recommended scenario to obtain the user-business modification evaluation matrix.
  • the modification submodule is specifically configured to:
  • the time impact characteristic in the recommendation scenario is: a characteristic that a positive influence of a time factor on a recommendation process increases with time.
  • the product evaluation value sequence of the user U a corresponding to the merchant S k in the first user-merchant initial evaluation matrix The corresponding network behavior occurs in a time series of The time-influence characteristic of the correction sub-module according to the recommended scenario and the time sequence of the network behavior occurrence Calculating the product evaluation value sequence Corresponding time correction factor sequence
  • T now indicates the current time
  • T period represents a time period in the recommended scenario
  • the correction sub-module is in the sequence of the commodity evaluation value Time correction factor sequence corresponding thereto Multiplying to obtain a modified evaluation value of the user U a corresponding to the merchant S k in the user-merchant correction evaluation matrix is specifically configured as:
  • v' ak represents a modified evaluation value of the user U a corresponding to the merchant S k .
  • the acquiring module is specifically configured to:
  • the recommendation module is specifically configured to:
  • the merchant-based collaborative filtering algorithm is used to recommend the merchant to the to-be-recommended user based on the revised evaluation value of the at least one user corresponding to the at least one candidate merchant.
  • the determining the mold body is configured to: acquire the at least one candidate merchant from the set of merchants according to the location of the user to be recommended.
  • Embodiments of the present disclosure also provide an electronic device, including: a memory and a processor; the memory is configured to store one or more computer instructions, wherein the one or more computer instructions are executable by the processor The steps in the recommended method provided by the above embodiments.
  • the embodiment of the present disclosure further provides a computer readable storage medium storing a computer program, the computer program causing a computer to perform the steps in the recommendation method provided by the above embodiments.
  • the initial evaluation value of the user is corrected, and then the recommendation is based on the corrected evaluation value, and the recommendation process is considered.
  • the purpose of the time factor is to improve the accuracy of the recommendation results.
  • FIG. 1 is a schematic flowchart diagram of a recommendation method according to an embodiment of the present disclosure
  • FIG. 2 is a schematic flowchart of a recommendation method according to another embodiment of the present disclosure.
  • 2b is a schematic diagram of an implementation form of a user-business modification evaluation matrix according to another embodiment of the present disclosure
  • 2c is a schematic diagram of another implementation form of a user-business modification evaluation matrix according to another embodiment of the present disclosure.
  • FIG. 3 is a schematic flowchart of a pre-built user-business modification evaluation matrix according to another embodiment of the present disclosure
  • FIG. 3b is a schematic diagram of an implementation form of a first user-commodity evaluation matrix according to another embodiment of the present disclosure
  • 3c is a schematic diagram of another implementation form of a first user-commodity evaluation matrix according to another embodiment of the present disclosure.
  • FIG. 3 is a schematic diagram of a process for converting a first user-commodity evaluation matrix from a user-item dimension to a user-business dimension according to another embodiment of the present disclosure
  • FIG. 4 is a schematic flowchart diagram of a recommendation method according to another embodiment of the present disclosure.
  • FIG. 5 is a schematic flowchart of a recommendation method according to another embodiment of the present disclosure.
  • FIG. 6 is a schematic flowchart of a recommendation method according to another embodiment of the present disclosure.
  • FIG. 7 is a schematic structural diagram of a recommendation apparatus according to another embodiment of the present disclosure.
  • FIG. 8 is a schematic structural diagram of a recommendation apparatus according to another embodiment of the present disclosure.
  • FIG. 1 is a schematic flowchart diagram of a recommendation method according to an embodiment of the present disclosure. As shown in FIG. 1, the method includes:
  • a user who needs to recommend content to it is referred to as a user to be recommended.
  • the content that needs to be recommended to the user will vary.
  • the embodiments of the present disclosure are mainly applicable to an application scenario with a plurality of merchants, and the merchant is recommended to the user in order to facilitate the user to purchase goods from a suitable merchant.
  • the recommendation method provided by the embodiment of the present disclosure can be applied to a shopping app provided by each major e-commerce, or a take-out app or the like.
  • the user to be recommended is first determined, and at least one candidate merchant is determined according to the user to be recommended.
  • the user to be recommended may be any user, such as an old user, a new user, or a potential user who may be a mobile Internet App.
  • all the merchants in the merchant set can be used as the candidate merchants.
  • the method for determining the candidate merchants is simple and efficient.
  • the determined number of candidate merchants is large, and the coverage is relatively comprehensive, which is beneficial to recommending to the user. A more suitable business.
  • At least one candidate merchant may be obtained from the merchant set according to the location of the user to be recommended. For example, a merchant located near the user to be recommended can be selected, which can reduce the number of candidate merchants, reduce the amount of calculation, save computing resources, and improve overall recommendation efficiency.
  • At least one merchant located within a specified range of the user to be recommended may be selected from the merchant set according to the location of the user to be recommended as the candidate merchant.
  • at least one merchant that is at a specified distance from the user to be recommended may be selected from the merchant set according to the location of the user to be recommended as the candidate merchant.
  • the merchant allows the user to evaluate the product and/or the goods it provides.
  • Users purchase goods from a certain merchant, and after using or consuming the goods, they generally evaluate the goods and/or merchants provided by the merchants.
  • the evaluation method of the products provided by the merchant or the merchant may be different.
  • a star icon is generally provided to the user, and the user selects the corresponding star icon to give the product provided by the merchant and/or the merchant, such as five stars, Samsung, etc.;
  • the user is provided with a text input box, and the user inputs text to evaluate the products provided by the merchant or the merchant.
  • the star rating icon selected by the user and the text input by the user can be comprehensively considered to determine the initial evaluation value of the user to the merchant.
  • the initial evaluation value of the candidate merchant for the user can be reflected by the user's evaluation of the product provided by the candidate merchant and/or the candidate merchant.
  • the merchant after obtaining the initial evaluation value of the candidate merchant by the user, the merchant is generally recommended from the candidate merchant to the user based on the initial evaluation value of the candidate merchant.
  • the initial evaluation value may be corrected according to the time influence characteristic conforming to the recommended scenario, and then the merchant is recommended to the user based on the modified evaluation value.
  • a modified evaluation value of at least one user corresponding to each of the at least one candidate merchant may be obtained, and the modified evaluation value is obtained by modifying the initial evaluation value according to the time influence characteristic in the recommended scenario.
  • At least one user may purchase or consume the commodity at the candidate merchant and make an evaluation, and then at least one of the commodities purchased at the candidate merchant and the evaluation may be determined.
  • User further, obtaining a revised evaluation value of the candidate merchant for each of the at least one user.
  • the corrected evaluation value of the at least one user corresponding to the candidate merchant actually refers to the correction of the candidate merchant by at least one user who has evaluated the candidate merchant and/or the commodity provided by the candidate merchant. Evaluation value.
  • the modified evaluation value of one candidate merchant is obtained by correcting the initial evaluation value of the candidate merchant according to the time influence characteristic in the recommendation scenario.
  • the initial evaluation value of the candidate merchant for the candidate merchant may be embodied by the user's evaluation of the candidate merchant and/or the commodity provided by the candidate merchant.
  • the initial evaluation value of the candidate merchant for the user may be directly expressed as the evaluation value of the at least one commodity provided by the user for the candidate merchant, or may also be expressed as the numerical processing result of the evaluation value of the at least one commodity provided by the candidate for the candidate merchant. Or, it can also be directly expressed as the user's evaluation value for the candidate merchant, and so on.
  • the impact characteristics of time factors will be different. For example, in some recommended scenarios, the impact of time factors on the recommendation process will gradually diminish over time. For example, in other recommended scenarios, the impact of time factors on the recommendation process will gradually increase over time. For another example, in some recommended scenarios, the influence of the time factor on the recommendation process will first weaken and then strengthen over time, or first, strengthen and then weaken.
  • the merchant is recommended to the user to be recommended based on the revised evaluation value of the at least one user corresponding to each of the at least one candidate merchant.
  • the embodiment does not limit the recommendation method used when recommending the merchant to the user to be recommended.
  • the method of recommending the merchant to the user to be recommended is based on the revised evaluation value of at least one user corresponding to the at least one candidate merchant. Disclosed embodiments.
  • the initial evaluation value of the user to the merchant is corrected, and then the recommendation is performed based on the revised evaluation value, and the purpose of the recommendation process considering the time factor is achieved. Thereby improving the accuracy of the recommendation results.
  • FIG. 2 is a schematic flowchart of a recommendation method according to another embodiment of the present disclosure. As shown in Figure 2a, the method includes:
  • the user-business-correction evaluation matrix includes a modified evaluation value of at least one user corresponding to each of the plurality of merchants in the system, and the modified evaluation value is a time-impacting characteristic according to the recommended scenario. Corrected the initial evaluation value.
  • the user-business correction evaluation matrix is pre-built to provide conditions for the recommendation process, such as step 200, before the recommendation is made.
  • the user-business correction evaluation matrix includes correction evaluation values of at least one user corresponding to each of the plurality of merchants in the system.
  • the revised evaluation value of the at least one user corresponding to the merchant actually refers to the modified evaluation of the merchant by at least one user who has evaluated the commodity provided by the merchant and/or the merchant. value.
  • the modified evaluation value of one candidate merchant is obtained by correcting the initial evaluation value of the candidate merchant according to the time influence characteristic in the recommendation scenario.
  • the initial evaluation value of the candidate merchant for the candidate merchant may be embodied by the user's evaluation of the candidate merchant and/or the commodity provided by the candidate merchant.
  • the initial evaluation value of the candidate merchant for the user may be directly expressed as the evaluation value of the at least one commodity provided by the user for the candidate merchant, or may also be expressed as the numerical processing result of the evaluation value of the at least one commodity provided by the candidate for the candidate merchant. Or, it can also be directly expressed as the user's evaluation value for the candidate merchant, and so on.
  • the above system refers to an application system including the above plurality of merchants.
  • the application system includes a server, a client on the merchant side, and a client on the user side.
  • the plurality of merchants may be all merchants in the application system, or may be some merchants in the application system.
  • the merchant is recommended to the user to be recommended, as in steps 201-203.
  • the user to be recommended is first determined, and according to the user to be recommended, at least one candidate merchant is determined from the plurality of merchants included in the user-business modification evaluation matrix, that is, the plurality of merchants includes at least one candidate merchant.
  • the user to be recommended may be any user, such as an old user, a new user, or a potential user who may be a mobile Internet App.
  • all the merchants in the plurality of merchants can be used as candidate merchants, and the method for determining the candidate merchants is simple and efficient; in addition, the determined number of candidate merchants is large, and the coverage is relatively comprehensive, which is beneficial to the user. Recommend a more suitable business.
  • At least one candidate merchant may be obtained from multiple merchants according to the location of the user to be recommended. For example, a merchant located near the user to be recommended can be selected, which can reduce the number of candidate merchants, reduce the amount of calculation, save computing resources, and improve overall recommendation efficiency.
  • step 201 in step 202, the revised evaluation value of at least one user corresponding to each of the at least one candidate merchant is obtained from the pre-built user-business modification evaluation matrix.
  • an implementation form of the evaluation matrix is modified for the user-business.
  • the user-business correction evaluation matrix includes a user identification, a merchant identification, and a revised evaluation value.
  • the user identifier constitutes a row in the user-businessman correction evaluation matrix
  • the merchant identifier constitutes a column in the user-businessman correction evaluation matrix
  • the modified evaluation value constitutes an element value in the user-businessman correction evaluation matrix.
  • the user-business correction evaluation matrix includes a user identification, a merchant identification, and a revised evaluation value.
  • the user identifier constitutes a column in the user-businessman correction evaluation matrix
  • the merchant identifier constitutes a row in the user-businessman correction evaluation matrix
  • the modified evaluation value constitutes an element value in the user-businessman correction evaluation matrix.
  • the above merchant identifier may be any information that can uniquely identify the merchant, such as the merchant name, the merchant ID, and the like.
  • the above user identifier may be any information that can uniquely identify the user, such as a user name, a user ID, and the like.
  • the revised evaluation value may be a specific value, such as 5 points, 3 points or 1 point; or, the revised evaluation value may also be some non-numeric information, such as a gold merchant, a silver merchant, a good reputation, a five-star service, Any discriminating information such as a three-star service.
  • the identifiers of the candidate merchants in the at least one candidate merchant can be matched in the user-business-correction evaluation matrix to obtain the user identifier corresponding to the merchant identifier in the matching. And determining, as the identifier of the user corresponding to each candidate merchant; and obtaining the corrected evaluation value determined by the matching merchant identifier and the corresponding user identifier, as the corrected evaluation value of each candidate merchant from the corresponding user.
  • the recommendation process of the embodiment it is not necessary to calculate the correction evaluation value of at least one user corresponding to each candidate merchant in real time, but to obtain at least one user corresponding to each candidate merchant directly based on the pre-built user-business modification evaluation matrix.
  • the revised evaluation value is more efficient and is conducive to improving the overall recommendation efficiency.
  • step 202 proceeding to step 203, the merchant is recommended to the user to be recommended based on the revised evaluation value of the at least one user corresponding to each of the at least one candidate merchant.
  • the embodiment does not limit the recommendation method used when recommending the merchant to the user to be recommended.
  • the method of recommending the merchant to the user to be recommended is based on the revised evaluation value of at least one user corresponding to the at least one candidate merchant. Disclosed embodiments.
  • the initial evaluation value of the user to the merchant is modified according to the time influence characteristic of the recommended scenario, and the user-business correction evaluation matrix is constructed; when the recommendation is made, directly from the user-business Correcting the evaluation evaluation value of the at least one user corresponding to each candidate merchant in the evaluation matrix, the efficiency is high, and then recommending the merchant based on the modified evaluation value of at least one user corresponding to each candidate merchant, thereby achieving the purpose of considering the time factor in the recommendation process, thereby Improve the accuracy of recommendation results and improve overall recommendation efficiency.
  • a process for constructing a user-business correction evaluation matrix in advance includes:
  • the network behavior information herein includes user evaluation information for the product, but is not limited thereto. Based on the user's evaluation information on the product, the user's evaluation value for the product can be obtained.
  • the first user-commodity evaluation matrix includes a user identification, a product identification, and an evaluation value of the user for the product.
  • the user identifier is a row in the first user-commodity evaluation matrix
  • the product identifier is a column in the first user-commodity evaluation matrix
  • the user's evaluation value for the commodity is an element value in the first user-commodity evaluation matrix.
  • the first user-commodity evaluation matrix includes a user identification, a product identification, and an evaluation value of the user for the product.
  • the user identifier is a column in the first user-commodity evaluation matrix
  • the product identifier is a row in the first user-commodity evaluation matrix
  • the user's evaluation value for the commodity is an element value in the first user-commodity evaluation matrix.
  • the embodiment of the present disclosure no longer recommends the product to the user as in the conventional recommendation method, but Using the relationship between the product and the merchant, the user's evaluation of the product is aggregated into a user's evaluation of the merchant, and the merchant is recommended to the user based on the user's evaluation of the merchant.
  • the relationship between the commodity and the merchant may be expressed as: the merchant provides the commodity for the user to consume, the merchant itself contains the attribute of the commodity, and the behavior of the user purchasing or consuming the commodity at the merchant may reflect the implicit interest of the user to the merchant.
  • the user's implicit interest in the merchant can be reflected in the number of purchases or consumption of goods at the merchant, the amount of consumption and evaluation.
  • the first user-commodity evaluation matrix needs to be switched from the user-item dimension to the user-merchant dimension.
  • Figure 3d it is a schematic diagram of the process of dimension conversion.
  • a dimension conversion manner may be: first, from the perspective of the merchant, obtaining the products belonging to the same merchant from the first user-commodity evaluation matrix; and then, according to the user classification, the products belonging to the same merchant are The same user summarizes the evaluation values of different commodities to form a product evaluation value sequence of the same user, thereby obtaining a product evaluation value sequence of at least one user corresponding to each of the plurality of merchants, that is, a first user-merchant initial evaluation matrix.
  • the first user-merchant initial evaluation matrix is modified according to the time influence characteristic in the recommended scenario to obtain a user-business correction evaluation matrix.
  • the first user-merchant initial evaluation matrix may be modified according to the time impact characteristic under the recommendation scenario to obtain a user-business modification evaluation matrix.
  • the first user-merchant initial evaluation matrix is modified according to the time influence characteristic in the recommended scenario, and the process of correcting the element value in the first user-merchant initial evaluation matrix according to the time influence characteristic in the recommended scenario is substantially .
  • an implementation manner of the foregoing step 2003 includes:
  • each product evaluation value sequence corresponds to one user and one merchant; for the user, at least one commodity is purchased or consumed at the merchant, and at least one commodity is evaluated.
  • the evaluation value of the at least one product by the user is summarized, and the product evaluation value sequence corresponding to the user is obtained.
  • each product evaluation value sequence includes an evaluation value of the product evaluation value sequence corresponding to at least one product provided by the user to the merchant corresponding to the product evaluation value sequence; correspondingly, a product evaluation value sequence also corresponds to a network behavior occurrence
  • the time series, the time sequence of occurrence of the network behavior is also the time when the user purchases or consumes the at least one commodity at the merchant.
  • its time-influence characteristic is the characteristic that the positive influence of the time factor on the recommendation process increases with time.
  • time For example, in a takeaway application scenario, users want to eat differently for dinner and lunch, eat differently today and yesterday, and eat differently on weekends and in peacetime. It can be seen that in this scenario, the user is generally not interested in what he has eaten recently, but as time goes by, the user will gradually become fresh after eating something he has eaten before, and eat it before the distance. The longer the time, the stronger the freshness will be, so it is necessary to recommend the previously consumed products to the user in a timely manner. This is the positive effect of the time factor on the recommendation process over time.
  • a commodity may be expressed as n ⁇ i 1, i 2, .. ., i n >
  • the sequence of commodity evaluation values of the user U a corresponding to the merchant S k in the first user-merchant initial evaluation matrix can be expressed as
  • the time series of the network behavior occurrence corresponding to the product evaluation value sequence can be expressed as That is, the user U a is at the moment S consumer goods merchant at k i 1, i 1 is formed and product evaluation value Accordingly, the user U a is at the moment S consumer goods merchant at k i 2, and i is formed product evaluation value 2 ..., and so on, user U a at the moment S consumer goods merchant at k i n, i n and product evaluation value is formed
  • T indicates the current time.
  • the T period indicates the time period in the recommended scenario, which is a preset value or an empirical value. For example, in a take-out scenario, the value of the time period may be 5, 7, 10, or 14 or the like. Indicates the evaluation value The corresponding network behavior occurs.
  • the product evaluation value sequence may be adopted by the following formula (2) Time correction factor sequence corresponding thereto Multiply to obtain a modified evaluation value of the user U a corresponding to the merchant S k in the user-merchant correction evaluation matrix.
  • the modified evaluation value of the user U a corresponding to the merchant S k is also the modified evaluation value of the user U a to the merchant S k .
  • v' ak represents the modified evaluation value of the user U a corresponding to the merchant S k .
  • each element value in the first user-merchant initial evaluation matrix may be modified according to the time influence characteristic in the recommended scenario, thereby obtaining a user-business corrected evaluation value matrix.
  • the user-business modification evaluation matrix is constructed offline, so that when the recommendation is made online, the correction evaluation value of at least one user corresponding to the candidate merchant can be directly obtained, which is beneficial to improving the efficiency of online recommendation.
  • the correction evaluation value of at least one user corresponding to the candidate merchant is directly obtained from the offline, and the candidate business correspondence may be calculated in real time when the recommendation is made online. At least one user's revised evaluation value, and then based on this, recommend the merchant to the user to be recommended.
  • FIG. 4 is a schematic flowchart diagram of a recommendation method according to another embodiment of the present disclosure. As shown in FIG. 4, the method includes:
  • the second user-business initial evaluation matrix is modified according to the time impact characteristic in the recommended scenario to obtain a user-candidate merchant correction evaluation matrix, where the user-candidate merchant correction evaluation matrix includes at least one candidate merchant corresponding to each at least A user's revised evaluation value.
  • step 401 For the description of the corresponding steps in the foregoing embodiment, reference may be made to step 401, and details are not described herein again.
  • Steps 402-404 a process for generating a modified evaluation value of at least one user corresponding to at least one candidate merchant in real time online, the process of which is similar to the process of pre-establishing the user-merchant correction evaluation matrix as shown in FIG. 3a, the difference is only: data The set is not the same.
  • steps 402-404 reference may be made to the process of the embodiment shown in Figure 3a, which will not be described in detail herein.
  • the merchant may be recommended to the user to be recommended based on the revised evaluation value of the at least one user corresponding to the at least one candidate merchant. effect.
  • the merchant may be recommended to the user to be recommended based on the revised evaluation value of at least one user corresponding to each of the at least one candidate merchant.
  • a user-based collaborative filtering algorithm may be adopted, and the merchant is recommended to the user to be recommended based on the modified evaluation value of the at least one user corresponding to the at least one candidate merchant.
  • a merchant-based collaborative filtering algorithm may be adopted, and the merchant is recommended to the user to be recommended based on the modified evaluation value of the at least one user corresponding to the at least one candidate merchant.
  • FIG. 5 is a schematic flowchart diagram of a recommendation method according to another embodiment of the present disclosure. As shown in Figure 5, the method includes:
  • step 502 may be implemented by referring to the embodiment shown in FIG. 2a and FIG. 3a, or may be implemented by referring to steps 402-404 in the embodiment shown in FIG. 4, and details are not described herein.
  • the first user set may be composed of all users existing in the system, or may be composed of some users existing in the system, or may also be composed of at least one user corresponding to each of the at least one candidate merchant.
  • an acquisition method is: obtaining a similar user of the user to be recommended from among all users existing in the system.
  • Another way to obtain is to obtain similar users of the users to be recommended from some users existing in the system.
  • Another acquisition manner is: obtaining a similar user of the user to be recommended from at least one user corresponding to each of the at least one candidate merchant.
  • a cosine similarity calculation method or a Pearson similarity calculation method may be used to obtain a similar user of the user to be recommended from the first user set.
  • the Pearson similarity calculation method may be used to calculate the similarity between the user to be recommended and each user in the first user set, and the Pearson similarity calculation method has the advantage of high accuracy of similarity calculation.
  • the Pearson similarity between u a and u b can be calculated by the following formula (3).
  • M represents the number of intersecting u u B and a merchant
  • the merchant refers to the presence of the intersection of the intersection of a u and u B (e.g. evaluated both) of the merchant
  • v a, j Representing the evaluation value of u a for the merchant s j in the intersecting merchant
  • Representing the average evaluation value of u a , v b,j represents the evaluation value of u b for the merchant s j
  • w a, b represents the u b u Pearson a similarity.
  • the similarity is calculated only actually u a and u b u a merchant based on the intersection of and u b. In practical applications, the number of intersecting merchants of u a and u b may be small. For this reason, the similarity calculated by the above formula (3) can be corrected by the following formula (4) to obtain the corrected similarity.
  • w' a, b represents the corrected similarity of u a and u b .
  • F represents a constant, and its value may be determined according to a specific application scenario, and may be, for example, but not limited to: 10.
  • Influential factor when the number M of intersecting merchants of u a and u b is greater than or equal to the constant F, it means that the number of intersecting merchants is large, and it is not necessary to use the influence factor to correct, so the influence factor takes 1; otherwise, when u a When the number M of intersecting merchants with u b is less than the constant F, it indicates that the number of intersecting merchants is small, and the impact factor is needed to be corrected, and the influence factor is less than 1. Using this influence factor can improve the accuracy and accuracy of the similarity calculation results.
  • the similarity between the user to be recommended and each user in the first user set can be calculated, and thus can be based on the user to be recommended and each user in the first user set. Similarity, determining similar users of the users to be recommended from the first set of users. For example, at least one user with the highest similarity to the user to be recommended may be selected as the similar user of the user to be recommended. For another example, at least one user whose similarity to the user to be recommended is greater than a set threshold may be selected as a similar user of the user to be recommended.
  • the corrected evaluation value of each of the at least one candidate merchant to be recommended by the user to be recommended may be calculated according to the following formula (5).
  • p a,i represents a modified evaluation value of the candidate merchant u a to the candidate merchant s i of the at least one candidate merchant
  • s i ⁇ S, S represents a candidate merchant set composed of at least one candidate merchant.
  • U' represents a similar set of users consisting of similar users of the user to be recommended;
  • w' a, c represents the corrected similarity of the user u a to be recommended and the similar user u c in the similar set of users;
  • v c, i means similar A modified evaluation value of the similar user u c in the user set to the candidate merchant s i of the at least one candidate merchant; Represents the average rating of similar users u c in a similar set of users.
  • At least one candidate merchant whose correction evaluation value is the largest may be selected and recommended to the user to be recommended.
  • at least one candidate merchant whose correction evaluation value is greater than the set threshold may be selected to be recommended to the user to be recommended.
  • the initial evaluation value of the user is corrected, and then the recommendation is made based on the corrected evaluation value, thereby achieving the purpose of considering the time factor in the recommendation process, thereby improving The accuracy of the recommended results.
  • the user-based collaborative filtering algorithm utilizes the similarity between users to perform merchant recommendation, has good prediction accuracy, and can change with changes in data, and is more suitable for systems with frequent data updates.
  • FIG. 6 is a schematic flowchart diagram of a recommendation method according to another embodiment of the present disclosure. As shown in FIG. 6, the method includes:
  • At least one candidate merchant whose correction evaluation value is the largest may be selected and recommended to the user to be recommended.
  • at least one candidate merchant whose correction evaluation value is greater than the set threshold may be selected to be recommended to the user to be recommended.
  • the first set of merchants may be composed of at least one merchant, or may be composed of all merchants existing in the system, or may be composed of some merchants existing in the system.
  • one way of obtaining is: acquiring at least one candidate merchant's respective similar merchants in at least one candidate merchant.
  • Another method of obtaining is: acquiring, among all the merchants existing in the system, the similar merchants of at least one candidate merchant.
  • Another acquisition method is: acquiring, among some merchants existing in the system, similar merchants of at least one candidate merchant.
  • a cosine similarity calculation method or a Pearson similarity calculation method may be used to obtain a similar merchant of each candidate merchant from the first merchant set.
  • the cosine similarity calculation method or the Pearson similarity calculation method is used to calculate the similarity between each candidate merchant and each merchant in the first merchant set, and the user is calculated according to the above formula (3) and formula (4).
  • the process of the similarity is only that the user-related parameters in the formula (3) and the formula (4) need to be replaced with the relevant parameters of the merchants in the candidate merchant and the first merchant set.
  • the specific calculation process is not detailed here. Said.
  • the merchant-based collaborative filtering algorithm utilizes the similarity between the merchants to perform the merchant recommendation, which can solve the problem of data sparseness and is beneficial to improve the recommendation accuracy.
  • the execution bodies of the steps of the method provided by the foregoing embodiments may all be the same device, or the method may also be performed by different devices.
  • the execution body of steps 101 to 103 may be device A; for example, the execution body of steps 101 and 102 may be device A, the execution body of step 103 may be device B, and the like.
  • FIG. 7 is a schematic structural diagram of a recommendation apparatus according to another embodiment of the present disclosure.
  • the recommendation device includes a determination module 71, an acquisition module 72, and a recommendation module 73.
  • the determining module 71 is configured to determine at least one candidate merchant according to the user to be recommended.
  • the obtaining module 72 is configured to obtain a modified evaluation value of at least one user corresponding to each of the at least one candidate merchant, wherein the modified evaluation value is obtained by modifying the initial evaluation value according to the time influence characteristic in the recommended scenario.
  • the recommendation module 73 is configured to recommend the merchant to the user to be recommended based on the revised evaluation value of the at least one user corresponding to each of the at least one candidate merchant.
  • the obtaining module 72 is specifically configured to: obtain, from the pre-built user-business-correction evaluation matrix, a modified evaluation value of at least one user corresponding to each of the at least one candidate merchant; wherein the user- The merchant correction evaluation matrix includes a revised evaluation value of at least one user corresponding to each of the plurality of merchants in the system, and the plurality of merchants include the at least one candidate merchant.
  • the recommending apparatus further includes: a building module 74 configured to pre-build a user-business correction evaluation matrix.
  • an implementation structure of the building module 74 includes: a building sub-module 741, a dimension switching sub-module 742, and a correction sub-module 743.
  • the construction sub-module 741 is configured to construct a first user-commodity evaluation matrix based on network behavior information of the user for the product existing in the system.
  • the dimension switching sub-module 742 is configured to perform dimension switching on the first user-commodity evaluation matrix to obtain a first user-merchant initial evaluation matrix, where the first user-merchant initial evaluation matrix includes the plurality of merchants respectively corresponding to each A sequence of commodity evaluation values for at least one user.
  • the correction sub-module 743 is configured to modify the first user-merchant initial evaluation matrix according to the time impact characteristic in the recommendation scenario to obtain a user-business correction evaluation matrix.
  • the modification sub-module 743 is specifically configured to:
  • the product evaluation value sequence in the first user-merchant initial evaluation matrix is multiplied by the corresponding time correction factor sequence to obtain a user-business correction evaluation matrix.
  • the time impact characteristic in the recommendation scenario is: a characteristic that a positive influence of a time factor on a recommendation process increases with time.
  • the correction sub-module 743 can specifically calculate the product evaluation value sequence according to the above formula (1). Corresponding time correction factor sequence
  • the correction sub-module 743 may specifically sequence the commodity evaluation value according to the above formula (2). Time correction factor sequence corresponding thereto Multiply to obtain a modified evaluation value of the user U a corresponding to the merchant S k in the user-merchant correction evaluation matrix.
  • the obtaining module 72 is specifically configured to: construct a second user-commodity evaluation matrix based on network behavior information of the product provided by the user for the at least one candidate merchant, and the second user-commodity
  • the evaluation matrix performs dimension switching to obtain a second user-merchant initial evaluation matrix, where the second user-merchant initial evaluation matrix includes a sequence of commodity evaluation values of at least one user corresponding to at least one candidate merchant; according to the time in the recommended scenario
  • the impact characteristic corrects the second user-merchant initial evaluation matrix to obtain a user-candidate merchant correction evaluation matrix, where the user-candidate merchant correction evaluation matrix includes correction evaluation values of at least one user corresponding to the at least one candidate merchant respectively .
  • the recommendation module 73 is specifically configured to: use a user-based collaborative filtering algorithm to recommend a merchant to the to-be-recommended user based on a modified evaluation value of at least one user corresponding to each of the at least one candidate merchant Or using a merchant-based collaborative filtering algorithm to recommend a merchant to the to-be-recommended user based on the revised evaluation value of at least one user corresponding to each of the at least one candidate merchant.
  • the determining module 71 is specifically configured to: acquire at least one candidate merchant from the set of merchants according to the location of the user to be recommended.
  • the recommended device provided by the embodiment of the present disclosure may be used to perform the process provided by the foregoing method embodiment, and the specific working principle is not described herein.
  • the recommendation device in the recommendation process, considers the influence of the time factor, corrects the initial evaluation value of the user based on the time influence characteristic of the recommended scenario, and then performs recommendation based on the corrected evaluation value to reach the recommendation process.
  • the time factor to improve the accuracy of the recommendation results.
  • embodiments of the present disclosure can be provided as a method, system, or computer program product. Accordingly, the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment, or a combination of software and hardware aspects. Moreover, the present disclosure may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) including computer usable program code.
  • computer-usable storage media including but not limited to disk storage, CD-ROM, optical storage, etc.
  • the computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture comprising the instruction device.
  • the apparatus implements the functions specified in one or more blocks of a flow or a flow and/or block diagram of the flowchart.
  • These computer program instructions can also be loaded onto a computer or other programmable data processing device such that a series of operational steps are performed on a computer or other programmable device to produce computer-implemented processing for execution on a computer or other programmable device.
  • the instructions provide steps for implementing the functions specified in one or more of the flow or in a block or blocks of a flow diagram.
  • a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
  • processors CPUs
  • input/output interfaces network interfaces
  • memory volatile and non-volatile memory
  • the memory may include non-persistent memory, random access memory (RAM), and/or non-volatile memory in a computer readable medium, such as read only memory (ROM) or flash memory.
  • RAM random access memory
  • ROM read only memory
  • Memory is an example of a computer readable medium.
  • Computer readable media includes both permanent and non-persistent, removable and non-removable media.
  • Information storage can be implemented by any method or technology.
  • the information can be computer readable instructions, data structures, modules of programs, or other data.
  • Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory. (ROM), electrically erasable programmable read only memory (EEPROM), flash memory or other memory technology, compact disk read only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, Magnetic tape cartridges, magnetic tape storage or other magnetic storage devices or any other non-transportable media can be used to store information that can be accessed by a computing device.
  • computer readable media does not include temporary storage of computer readable media, such as modulated data signals and carrier waves.
  • embodiments of the present disclosure may be provided as a method, system, or computer program product. Accordingly, the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment or a combination of software and hardware aspects. Moreover, the present disclosure may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) including computer usable program code.
  • computer-usable storage media including but not limited to disk storage, CD-ROM, optical storage, etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Development Economics (AREA)
  • Databases & Information Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • General Engineering & Computer Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • Operations Research (AREA)
  • Tourism & Hospitality (AREA)
  • Quality & Reliability (AREA)
  • Human Resources & Organizations (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Game Theory and Decision Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种推荐方法、装置、电子设备以及存储介质,其中推荐方法包括:根据待推荐用户,确定至少一个候选商户(101);获取至少一个候选商户各自对应的至少一个用户的修正评价值,修正评价值是按照推荐场景下的时间影响特性对初始评价值修正得到的(102);基于至少一个候选商户各自对应的至少一个用户的修正评价值,向待推荐用户推荐商户(103)。从而可以提高推荐结果的准确性。

Description

推荐方法及装置 技术领域
本公开涉及互联网技术领域,尤其涉及一种推荐方法及装置。
背景技术
随着互联网技术的发展,个性化推荐服务在互联网行业得到越来越多的应用,例如在地图上查找附近的服务、在外卖应用中搜索附近的餐馆等。为了给用户推荐感兴趣的内容,例如服务或餐馆,除了使用地理位置之外,还可以收集和分析用户的浏览信息、历史订单和商品评价等行为数据,确定用户兴趣,进而基于用户兴趣做推荐。
相关技术中常见的基于用户兴趣的推荐方法主要包括:基于内容的推荐方法和协同过滤推荐(Collaborative Filtering Recommendation)方法。相关推荐方法或多或少都会存在一些缺陷,相关方法存在的缺陷最终会影响推荐结果,导致不能完成推荐或推荐不准确。
发明内容
本公开发明人对现有基于内容的推荐方法以及协同过滤的推荐方法进行了分析。
基于内容的推荐方法,其核心主要是采用自然语言处理、人工智能、概率统计和机器学习等技术进行内容过滤,努力发现用户的兴趣,向用户推荐与用户以前喜欢的商品相似的商品。
协同过滤推荐方法,其核心主要是分析用户兴趣,在用户群中找到该用户的相似用户,综合这些相似用户对某一信息的评价,形成该用户对此信息的喜好程度,据此向用户进行推荐。
上述推荐方法的性能已经很好,但依然存在无法完成推荐或推荐不 准确的问题。对此,本公开发明人进一步进行了研究分析,上述推荐方法考虑了地理位置、用户兴趣、用户行为等因素,但却忽略了时间对用户行为的影响。例如,用户希望晚餐和午餐吃得不一样、今天和昨天吃得不一样、周末和平时吃得也不一样。由此可见,在这种场景下,若采用现有推荐方法向用户推荐以前喜欢或吃过的东西反而不合适,为了用户饮食的多样性,应该向用户推荐近期未吃过的东西。另外,随着时间的推移,有些吃过的东西,又会变得有新鲜感,因此也需要适时推荐之前消费过的类似商品。
针对上述分析,本公开发明人提供一种推荐方法,其核心是:在推荐过程中,考虑时间因素的影响,基于符合推荐场景的时间影响特性,修正用户对商品的初始评价值,进而基于修正后的评价值进行推荐,达到推荐过程考虑时间因素的目的,从而提高推荐结果的准确性。
基于上述,本公开实施例提供一种推荐方法,包括:
根据待推荐用户,确定至少一个候选商户;
获取所述至少一个候选商户各自对应的至少一个用户的修正评价值,所述修正评价值是按照推荐场景下的时间影响特性对初始评价值修正得到的;
基于所述至少一个候选商户各自对应的至少一个用户的修正评价值,向所述待推荐用户推荐商户。
在一可选实施方式中,所述获取所述至少一个候选商户各自对应的至少一个用户的修正评价值,包括:
从预先构建的用户-商户修正评价矩阵中,获取所述至少一个候选商户各自对应的至少一个用户的修正评价值;
其中,所述用户-商户修正评价矩阵包括系统中多个商户各自对应的至少一个用户的修正评价值,所述多个商户包含所述至少一个候选商户。
在一可选实施方式中,预先构建用户-商户修正评价矩阵,包括:
基于系统中存在的用户针对商品的网络行为信息,构建第一用户-商品评价矩阵;
对所述第一用户-商品评价矩阵进行维度切换,以获得第一用户-商 户初始评价矩阵,所述第一用户-商户初始评价矩阵包括所述多个商户各自对应的至少一个用户的商品评价值序列;
根据所述推荐场景下的时间影响特性对所述第一用户-商户初始评价矩阵进行修正,以获得所述用户-商户修正评价矩阵。
在一可选实施方式中,所述根据所述推荐场景下的时间影响特性对所述第一用户-商户初始评价矩阵进行修正,以获得所述用户-商户修正评价矩阵,包括:
根据所述推荐场景下的时间影响特性以及所述第一用户-商户初始评价矩阵中各商品评价值序列对应的网络行为发生时间序列,计算所述第一用户-商户初始评价矩阵中各商品评价值序列对应的时间修正因子序列;
将所述第一用户-商户初始评价矩阵中各商品评价值序列与各自对应的时间修正因子序列相乘,以获得所述用户-商户修正评价矩阵。
在一可选实施方式中,所述推荐场景下的时间影响特性为:时间因素对推荐过程的正向影响随时间增长的特性。
在一可选实施方式中,对所述第一用户-商户初始评价矩阵中商户S k对应的用户U a的商品评价值序列
Figure PCTCN2017118777-appb-000001
其对应的网络行为发生时间序列为
Figure PCTCN2017118777-appb-000002
所述根据所述推荐场景下的时间影响特性以及所述网络行为发生时间序列
Figure PCTCN2017118777-appb-000003
计算所述商品评价值序列
Figure PCTCN2017118777-appb-000004
Figure PCTCN2017118777-appb-000005
对应的时间修正因子序列
Figure PCTCN2017118777-appb-000006
包括:
根据公式
Figure PCTCN2017118777-appb-000007
计算所述商品评价值序列
Figure PCTCN2017118777-appb-000008
对应的时间修正因子序列
Figure PCTCN2017118777-appb-000009
中的时间修正因子;
Figure PCTCN2017118777-appb-000010
表示所述商品评价值序列
Figure PCTCN2017118777-appb-000011
中评价值
Figure PCTCN2017118777-appb-000012
对应的时间修正因子,1≤j≤n,j和n是自然数;
T now表示当前时间;
T period表示所述推荐场景下的时间周期;
Figure PCTCN2017118777-appb-000013
表示评价值
Figure PCTCN2017118777-appb-000014
对应的网络行为发生时间。
在一可选实施方式中,将所述商品评价值序列
Figure PCTCN2017118777-appb-000015
与其对应的时间修正因子序列
Figure PCTCN2017118777-appb-000016
相乘,以获得所述用户-商户修正评价矩阵中商户S k对应的用户U a的修正评价值,包括:
根据公式
Figure PCTCN2017118777-appb-000017
计算所述商户S k对应的用户U a的修正评价值;
v′ ak表示所述商户S k对应的用户U a的修正评价值。
在一可选实施方式中,所述获取所述至少一个候选商户各自对应的至少一个用户的修正评价值,包括:
基于系统中存在的用户针对所述至少一个候选商户提供的商品的网络行为信息,构建第二用户-商品评价矩阵;
对所述第二用户-商品评价矩阵进行维度切换,以获得第二用户-商户初始评价矩阵,所述第二用户-商户初始评价矩阵包括所述至少一个候选商户各自对应的至少一个用户的商品评价值序列;
根据所述推荐场景下的时间影响特性对所述第二用户-商户初始评价矩阵进行修正,以获得用户-候选商户修正评价矩阵,所述用户-候选商户修正评价矩阵包括所述至少一个候选商户各自对应的至少一个用户的修正评价值。
在一可选实施方式中,所述基于所述至少一个候选商户各自对应的至少一个用户的修正评价值,向所述待推荐用户推荐商户,包括:
采用基于用户的协同过滤算法,基于所述至少一个候选商户各自对应的至少一个用户的修正评价值,向所述待推荐用户推荐商户;或者
采用基于商户的协同过滤算法,基于所述至少一个候选商户各自对应的至少一个用户的修正评价值,向所述待推荐用户推荐商户。
在一可选实施方式中,所述根据待推荐用户,确定至少一个候选商户,包括:
根据所述待推荐用户的位置,从商户集合中,获取所述至少一个候选商户。
相应地,本公开实施例还提供一种推荐装置,包括:
确定模块,被配置为根据待推荐用户,确定至少一个候选商户;
获取模块,被配置为获取所述至少一个候选商户各自对应的至少一个用户的修正评价值,所述修正评价值是按照推荐场景下的时间影响特性对初始评价值修正得到的;
推荐模块,被配置为基于所述至少一个候选商户各自对应的至少一个用户的修正评价值,向所述待推荐用户推荐商户。
在一可选实施方式中,所述获取模块具体被配置为:
从预先构建的用户-商户修正评价矩阵中,获取所述至少一个候选商户各自对应的至少一个用户的修正评价值;
其中,所述用户-商户修正评价矩阵包括系统中多个商户各自对应的至少一个用户的修正评价值,所述多个商户包含所述至少一个候选商户。
在一可选实施方式中,所述装置还包括:
构建模块,被配置为预先构建所述用户-商户修正评价矩阵;
所述构建模块包括:
构建子模块,被配置为基于系统中存在的用户针对商品的网络行为信息,构建第一用户-商品评价矩阵;
维度切换子模块,被配置为对所述第一用户-商品评价矩阵进行维度切换,以获得第一用户-商户初始评价矩阵,所述第一用户-商户初始评价矩阵包括所述多个商户各自对应的至少一个用户的商品评价值序列;
修正子模块,被配置为根据所述推荐场景下的时间影响特性对所述第一用户-商户初始评价矩阵进行修正,以获得所述用户-商户修正评价矩阵。
在一可选实施方式中,所述修正子模块具体被配置为:
根据所述推荐场景下的时间影响特性以及所述第一用户-商户初始 评价矩阵中各商品评价值序列对应的网络行为发生时间序列,计算所述第一用户-商户初始评价矩阵中各商品评价值序列对应的时间修正因子序列;
将所述第一用户-商户初始评价矩阵中各商品评价值序列与各自对应的时间修正因子序列相乘,以获得所述用户-商户修正评价矩阵。
在一可选实施方式中,所述推荐场景下的时间影响特性为:时间因素对推荐过程的正向影响随时间增长的特性。
在一可选实施方式中,对所述第一用户-商户初始评价矩阵中商户S k对应的用户U a的商品评价值序列
Figure PCTCN2017118777-appb-000018
其对应的网络行为发生时间序列为
Figure PCTCN2017118777-appb-000019
所述修正子模块在根据所述推荐场景下的时间影响特性以及所述网络行为发生时间序列
Figure PCTCN2017118777-appb-000020
计算所述商品评价值序列
Figure PCTCN2017118777-appb-000021
对应的时间修正因子序列
Figure PCTCN2017118777-appb-000022
时,具体被配置为:
根据公式
Figure PCTCN2017118777-appb-000023
计算所述商品评价值序列
Figure PCTCN2017118777-appb-000024
对应的时间修正因子序列
Figure PCTCN2017118777-appb-000025
中的时间修正因子;
Figure PCTCN2017118777-appb-000026
表示所述商品评价值序列
Figure PCTCN2017118777-appb-000027
中评价值
Figure PCTCN2017118777-appb-000028
对应的时间修正因子,1≤j≤n,j和n是自然数;
T now表示当前时间;
T period表示所述推荐场景下的时间周期;
Figure PCTCN2017118777-appb-000029
表示评价值
Figure PCTCN2017118777-appb-000030
对应的网络行为发生时间。
在一可选实施方式中,所述修正子模块在将所述商品评价值序列
Figure PCTCN2017118777-appb-000031
Figure PCTCN2017118777-appb-000032
与其对应的时间修正因子序列
Figure PCTCN2017118777-appb-000033
相乘,以获得所述用户-商户修正评价矩阵中商户S k对应的用户U a的修正评价值时,具 体被配置为:
根据公式
Figure PCTCN2017118777-appb-000034
计算所述商户S k对应的用户U a的修正评价值;
v′ ak表示所述商户S k对应的用户U a的修正评价值。
在一可选实施方式中,所述获取模块具体被配置为:
基于系统中存在的用户针对所述至少一个候选商户提供的商品的网络行为信息,构建第二用户-商品评价矩阵;
对所述第二用户-商品评价矩阵进行维度切换,以获得第二用户-商户初始评价矩阵,所述第二用户-商户初始评价矩阵包括所述至少一个候选商户各自对应的至少一个用户的商品评价值序列;
根据所述推荐场景下的时间影响特性对所述第二用户-商户初始评价矩阵进行修正,以获得用户-候选商户修正评价矩阵,所述用户-候选商户修正评价矩阵包括所述至少一个候选商户各自对应的至少一个用户的修正评价值。
在一可选实施方式中,所述推荐模块具体被配置为:
采用基于用户的协同过滤算法,基于所述至少一个候选商户各自对应的至少一个用户的修正评价值,向所述待推荐用户推荐商户;或者
采用基于商户的协同过滤算法,基于所述至少一个候选商户各自对应的至少一个用户的修正评价值,向所述待推荐用户推荐商户。
在一可选实施方式中,所述确定模具体被配置为:根据所述待推荐用户的位置,从商户集合中,获取所述至少一个候选商户。
本公开实施例还提供一种电子设备,包括:存储器和处理器;所述存储器用于存储一条或多条计算机指令,其中,所述一条或多条计算机指令被所述处理器执行时能够实现上述实施例提供的推荐方法中的步骤。
本公开实施例还提供一种存储有计算机程序的计算机可读存储介质,所述计算机程序使计算机执行上述实施例提供的推荐方法中的步骤。
在本公开实施例中,在推荐过程中,考虑时间因素的影响,基于符合推荐场景的时间影响特性,修正用户对商品的初始评价值,进而基于修正后的评价值进行推荐,达到推荐过程考虑时间因素的目的,从而提高推荐结果的准确性。
附图说明
此处所说明的附图用来提供对本公开的进一步理解,构成本公开的一部分,本公开的示意性实施例及其说明被配置为解释本公开,并不构成对本公开的不当限定。在附图中:
图1为本公开一实施例提供的推荐方法的流程示意图;
图2a为本公开另一实施例提供的推荐方法的流程示意图;
图2b为本公开另一实施例提供的用户-商户修正评价矩阵的一种实现形式的示意图;
图2c为本公开另一实施例提供的用户-商户修正评价矩阵的另一种实现形式的示意图;
图3a为本公开又一实施例提供的预先构建用户-商户修正评价矩阵的流程示意图;
图3b为本公开又一实施例提供的第一用户-商品评价矩阵的一种实现形式的示意图;
图3c为本公开又一实施例提供的第一用户-商品评价矩阵的另一种实现形式的示意图;
图3d为本公开又一实施例提供的第一用户-商品评价矩阵从用户-商品维度向用户-商户维度进行转换的过程示意图;
图4为本公开又一实施例提供的推荐方法的流程示意图;
图5为本公开又一实施例提供的推荐方法的流程示意图;
图6为本公开又一实施例提供的推荐方法的流程示意图;
图7为本公开又一实施例提供的推荐装置的结构示意图;
图8为本公开又一实施例提供的推荐装置的结构示意图。
具体实施方式
为使本公开的目的、技术方案和优点更加清楚,下面将结合本公开具体实施例及相应的附图对本公开技术方案进行清楚、完整地描述。显然,所描述的实施例仅是本公开一部分实施例,而不是全部的实施例。基于本公开中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都属于本公开保护的范围。
图1为本公开一实施例提供的推荐方法的流程示意图。如图1所示,所述方法包括:
101、根据待推荐用户,确定至少一个候选商户。
102、获取至少一个候选商户各自对应的至少一个用户的修正评价值,所述修正评价值是按照推荐场景下的时间影响特性对初始评价值修正得到的。
103、基于至少一个候选商户各自对应的至少一个用户的修正评价值,向待推荐用户推荐商户。
在移动互联网应用(Application,App)的使用过程中,经常因为各种原因需要向用户进行相关内容的推荐。为了便于描述,在本公开实施例中,将需要向其推荐内容的用户称之为待推荐用户。根据应用场景的不同,需要向用户推荐的内容也会有所不同。本公开实施例主要适用于具有众多商户的应用场景中,为便于用户从合适的商户选购商品,向用户推荐商户。举例说明,本公开实施例提供的推荐方法可应用于各大电商提供的购物类App,或者外卖类App等。
参见步骤101,在推荐过程中,首先确定待推荐用户,根据待推荐用户确定至少一个候选商户。待推荐用户可以是任何用户,例如可以是移动互联网App的老用户、新用户或者潜在用户。
可选地,可以将商户集合中的所有商户都作为候选商户,这种确定候选商户的方式简单、效率较高;另外,所确定的候选商户数量较多,覆盖面比较全,有利于向用户推荐出更加合适的商户。
可选地,可以根据待推荐用户的位置,从商户集合中,获取至少一个候选商户。例如,可以选择位于待推荐用户附近的商户,这样可以减少候选商户的数量,有利于减少计算量,节约计算资源,提高整体推荐效率。
例如,可以根据待推荐用户的位置,从商户集合中,选择位于待推荐用户指定范围内的至少一个商户,作为候选商户。或者,可以根据待推荐用户的位置,从商户集合中,选择与待推荐用户相距指定距离的至少一个商户,作为候选商户。
在实际应用过程中,为了提高服务质量和用户体验度,商户允许用户对其和/或其提供的商品进行评价。用户从某一商户处选购商品,在使用或消费商品后,一般会对商户提供的商品和/或商户进行评价。其中,根据应用场景的不同,用户对商户或商户提供的商品的评价方式也会有所不同。例如,在购物应用场景或外卖类应用场景中,一般会向用户提供星级图标,由用户选择相应星级图标给商户和/或商户提供的商品打分,例如五星、三星等;另外,还会向用户提供文本输入框,由用户输入文字对商户或商户提供的商品进行评价。
基于上述,可以综合考虑用户选择的星级图标和用户输入的文字,确定用户对商户的初始评价值。换句话说,用户对候选商户的初始评价值可通过用户对候选商户和/或候选商户提供的商品做出的评价来体现。
在现有技术中,在获得用户对候选商户的初始评价值之后,一般会基于该用户对候选商户的初始评价值,从候选商户中向用户推荐商户。但在本实施例中,考虑到时间因素在推荐过程中的影响,可以根据符合推荐场景的时间影响特性对初始评价值进行修正,进而基于修正评价值向用户推荐商户。
参见步骤102,可以获取至少一个候选商户各自对应的至少一个用户的修正评价值,所述修正评价值是按照推荐场景下的时间影响特性对初始评价值修正得到的。
对至少一个候选商户中的任一候选商户,可能有至少一个用户在该候选商户处选购或消费商品并做出评价,则可以确定在该候选商户处选 购商品并且做出评价的至少一个用户;进一步,获取至少一个用户中每个用户对该候选商户的修正评价值。从候选商户的角度来看,该候选商户对应的至少一个用户的修正评价值实际上是指对该候选商户和/或该候选商户提供的商品做过评价的至少一个用户对该候选商户的修正评价值。
其中,一个用户对一个候选商户的修正评价值是按照推荐场景下的时间影响特性对该用户对该候选商户的初始评价值进行修正得到的。其中,该用户对该候选商户的初始评价值可通过该用户对该候选商户和/或该候选商户提供的商品做出的评价来体现。例如,用户对候选商户的初始评价值可以直接表示为用户对该候选商户提供的至少一个商品的评价值,或者也可以表示为用户对该候选商户提供的至少一个商品的评价值的数值处理结果,或者也可以直接表示为用户对该候选商户的评价值,等等。
其中,不同推荐场景,时间因素的影响特性会有所不同。例如,在一些推荐场景中,时间因素对推荐过程的影响会随着时间的推移而逐渐削弱。例如,在另一些推荐场景中,时间因素对推荐过程的影响会随着时间的推移而逐渐加强。又例如,在又一些推荐场景中,时间因素对推荐过程的影响会先随着时间的推移先削弱再加强,又或者,先加强再削弱。
值得说明的是,对不同候选商户来说,对其和/或其提供的商品做出评价的用户可能相同,也可能不同。
继续参见步骤103,基于至少一个候选商户各自对应的至少一个用户的修正评价值,向待推荐用户推荐商户。
本实施例并不限定向待推荐用户推荐商户时采用的推荐方式,凡是以至少一个候选商户各自对应的至少一个用户的修正评价值为基础,可以向待推荐用户推荐商户的方式均适用于本公开实施例。
在本实施例的推荐过程中,考虑时间因素的影响,基于符合推荐场景的时间影响特性,修正用户对商户的初始评价值,进而基于修正评价值进行推荐,达到推荐过程考虑时间因素的目的,从而提高推荐结果的 准确性。
图2a为本公开另一实施例提供的推荐方法的流程示意图。如图2a所示,所述方法包括:
200、预先构建用户-商户修正评价矩阵,所述用户-商户修正评价矩阵包括系统中多个商户各自对应的至少一个用户的修正评价值,所述修正评价值是按照推荐场景下的时间影响特性对初始评价值修正得到的。
201、根据待推荐用户,从所述多个商户中,确定至少一个候选商户。
202、从预先构建的用户-商户修正评价矩阵中,获取至少一个候选商户各自对应的至少一个用户的修正评价值。
203、基于至少一个候选商户各自对应的至少一个用户的修正评价值,向待推荐用户推荐商户。
在本实施例中,在进行推荐之前,预先构建用户-商户修正评价矩阵,为推荐过程提供条件,如步骤200。该用户-商户修正评价矩阵包括系统中多个商户各自对应的至少一个用户的修正评价值。对多个商户中的任一商户来说,该商户对应的至少一个用户的修正评价值实际上是指对该商户和/或商户提供的商品进行过评价的至少一个用户对该商户的修正评价值。其中,一个用户对一个候选商户的修正评价值是按照推荐场景下的时间影响特性对该用户对该候选商户的初始评价值进行修正得到的。其中,该用户对该候选商户的初始评价值可通过该用户对该候选商户和/或该候选商户提供的商品做出的评价来体现。例如,用户对候选商户的初始评价值可以直接表示为用户对该候选商户提供的至少一个商品的评价值,或者也可以表示为用户对该候选商户提供的至少一个商品的评价值的数值处理结果,或者也可以直接表示为用户对该候选商户的评价值,等等。
上述系统是指包含上述多个商户的应用系统。在本实施例中,应用系统包含服务端、商户一侧的客户端以及用户一侧的客户端。所述多个商户可以是应用系统中的全部商户,也可以是应用系统中的部分商户。
在推荐过程中,基于预先构建的用户-商户修正评价矩阵,向待推荐 用户推荐商户,如步骤201-203。
在步骤201中,首先确定待推荐用户,根据待推荐用户,从用户-商户修正评价矩阵包含的多个商户中,确定至少一个候选商户,即,所述多个商户包含至少一个候选商户。待推荐用户可以是任何用户,例如可以是移动互联网App的老用户、新用户或者潜在用户。
可选地,可以将多个商户中的所有商户都作为候选商户,这种确定候选商户的方式简单、效率较高;另外,所确定的候选商户数量较多,覆盖面比较全,有利于向用户推荐出更加合适的商户。
可选地,可以根据待推荐用户的位置,从多个商户中,获取至少一个候选商户。例如,可以选择位于待推荐用户附近的商户,这样可以减少候选商户的数量,有利于减少计算量,节约计算资源,提高整体推荐效率。
接续于步骤201,在步骤202中,从预先构建的用户-商户修正评价矩阵中,获取至少一个候选商户各自对应的至少一个用户的修正评价值。
可选地,如图2b所示,为用户-商户修正评价矩阵的一种实现形式。如图2b所示,用户-商户修正评价矩阵包含用户标识、商户标识以及修正评价值。其中,用户标识构成用户-商户修正评价矩阵中的行,商户标识构成用户-商户修正评价矩阵中的列,修正评价值构成用户-商户修正评价矩阵中的元素值。
可选地,如图2c所示,为用户-商户修正评价矩阵的另一种实现形式。如图2c所示,用户-商户修正评价矩阵包含用户标识、商户标识以及修正评价值。其中,用户标识构成用户-商户修正评价矩阵中的列,商户标识构成用户-商户修正评价矩阵中的行,修正评价值构成用户-商户修正评价矩阵中的元素值。
在图2b或图2c中,用户集合表示为U,对用户u a来说,存在以下所属关系:u a∈U={u 1,u 2,…,u N},N表示用户个数;商户集合表示为S,对商品s j来说,存在以下所属关系:s j∈S={s 1,s 2,…,s K},K表示商户个数。
上述商户标识可以是商户名称、商户ID等任何能够唯一标识商户的信息。相应地,上述用户标识可以是用户名称、用户ID等任何能够唯一标识用户的信息。修正评价值可以是具体的数值,例如5分、3分或1分;或者,修正评价值也可以是一些非数值型的信息,例如金牌商户、银牌商户、口碑较好、五星级服务、三星级服务等任何具有区分度的信息。
基于图2b或图2c所示的用户-商户修正评价矩阵,可以将至少一个候选商户中各候选商户的标识分别在用户-商户修正评价矩阵中进行匹配,获取匹配中的商户标识对应的用户标识,作为各候选商户对应的用户的标识;进而获取匹配中的商户标识以及对应的用户标识所确定的修正评价值,作为各候选商户自对应的用户的修正评价值。
在本实施例的在推荐过程中,无需实时计算各候选商户对应的至少一个用户的修正评价值,而是基于预先构建的用户-商户修正评价矩阵,直接从中获取各候选商户对应的至少一个用户的修正评价值,效率较高,有利于提高整体推荐效率。
接续于步骤202,继续参见步骤203,基于至少一个候选商户各自对应的至少一个用户的修正评价值,向待推荐用户推荐商户。
本实施例并不限定向待推荐用户推荐商户时采用的推荐方式,凡是以至少一个候选商户各自对应的至少一个用户的修正评价值为基础,可以向待推荐用户推荐商户的方式均适用于本公开实施例。
在本实施例中,考虑时间因素的影响,预先根据符合推荐场景的时间影响特性,修正用户对商户的初始评价值,构建出用户-商户修正评价矩阵;在进行推荐时,直接从用户-商户修正评价矩阵中获取各候选商户对应的至少一个用户的修正评价值,效率较高,进而基于各候选商户对应的至少一个用户的修正评价值进行商户推荐,达到推荐过程考虑时间因素的目的,从而提高推荐结果的准确性,提高整体推荐效率。
在上述实施例或下述实施例中,一种预先构建用户-商户修正评价矩阵的流程,如图3a所示,包括:
2001、基于系统中存在的用户针对商品的网络行为信息,构建第一 用户-商品评价矩阵。
这里的网络行为信息包括用户针对商品的评价信息,但不限于此。基于用户对商品的评价信息,可以获得用户对商品的评价值。
如图3b所示,为第一用户-商品评价矩阵的一种实现形式。在图3b中,第一用户-商品评价矩阵包括用户标识、商品标识以及用户对商品的评价值。其中,用户标识为第一用户-商品评价矩阵中的行,商品标识为第一用户-商品评价矩阵中的列,用户对商品的评价值为第一用户-商品评价矩阵中的元素值。
如图3c所示,为第一用户-商品评价矩阵的另一种实现形式。在图3c中,第一用户-商品评价矩阵包括用户标识、商品标识以及用户对商品的评价值。其中,用户标识为第一用户-商品评价矩阵中的列,商品标识为第一用户-商品评价矩阵中的行,用户对商品的评价值为第一用户-商品评价矩阵中的元素值。
在图3b或图3c中,用户集合表示为U,对用户u a来说,存在以下所属关系:u a∈U={u 1,u 2,…,u N},N表示用户个数;商品集合表示为I,对商品i j来说,存在以下所属关系:i j∈I={i 1,i 2,…,i M},M表示商品个数。
2002、对第一用户-商品评价矩阵进行维度切换,以获得第一用户-商户初始评价矩阵,所述第一用户-商户初始评价矩阵包括所述多个商户各自对应的至少一个用户的商品评价值序列。
在本公开实施例提供的推荐方法中,考虑到商品种类繁多,在基于商品进行推荐时容易存在数据比较稀疏的问题,所以本公开实施例不再像传统推荐方法那样向用户推荐商品,而是利用商品和商户的关系,将用户对商品的评价聚合成一个用户对商户的评价,基于用户对商户的评价向用户推荐商户。其中,商品与商户之间的关系可表现为:商户提供商品供用户消费,商户本身包含商品的属性,而用户在商户选购或消费商品的行为可以反映用户对商户的隐含兴趣。用户对商户的隐含兴趣可以体现为在商户处的选购或消费商品的次数、消费金额和评价等信息。
在本实施例中,需要将第一用户-商品评价矩阵从用户-商品维度切换为用户-商户维度。如图3d所示,为维度转换的过程示意图。可选地,一种维度转换的方式可以是:先从商户的角度出发,从第一用户-商品评价矩阵中获取属于同一商户的商品;然后对属于同一商户的商品,再按照用户区分,将同一用户对不同商品的评价值进行汇总,形成同一用户的商品评价值序列,从而获得多个商户各自对应的至少一个用户的商品评价值序列,即第一用户-商户初始评价矩阵。
2003、根据推荐场景下的时间影响特性对第一用户-商户初始评价矩阵进行修正,以获得用户-商户修正评价矩阵。
在获得第一用户-商户初始评价矩阵之后,可以根据推荐场景下的时间影响特性对第一用户-商户初始评价矩阵进行修正,以获得用户-商户修正评价矩阵。其中,根据推荐场景下的时间影响特性对第一用户-商户初始评价矩阵进行修正,实质上是根据推荐场景下的时间影响特性对第一用户-商户初始评价矩阵中的元素值进行修正的过程。
可选地,上述步骤2003的一种实施方式,包括:
根据推荐场景下的时间影响特性以及第一用户-商户初始评价矩阵中各商品评价值序列对应的网络行为发生时间序列,计算第一用户-商户初始评价矩阵中各商品评价值序列对应的时间修正因子序列;然后将第一用户-商户初始评价矩阵中各商品评价值序列与各自对应的时间修正因子序列相乘,以获得用户-商户修正评价矩阵。
在第一用户-商户初始评价矩阵中,每个商品评价值序列对应一个用户与一个商户;对该用户来说,会在该商户处选购或消费至少一个商品并对至少一个商品进行评价,这样该用户对至少一个商品的评价值,对这些评价值进行汇总,可以得到该商户对应于该用户的商品评价值序列。换句话说,每个商品评价值序列包含该商品评价值序列对应用户对该商品评价值序列对应商户提供的至少一个商品的评价值;相应地,一个商品评价值序列也会对应一个网络行为发生时间序列,所述网络行为发生时间序列也就是该用户在该商户处选购或消费所述至少一个商品的时间。
用户在商户处选购或消费商品的行为存在时间上的先后关系,这种先后关系可以体现出用户在商户处选购或消费商品的行为在时间上具有的周期性习惯,这种周期性习惯会对推荐过程造成影响,也就是时间因素的影响。
在一种具体推荐场景下,其时间影响特性为:时间因素对推荐过程的正向影响随时间增长的特性。例如,在外卖应用场景中,用户希望晚餐和午餐吃得不一样、今天和昨天吃得不一样、周末和平时吃得也不一样。由此可见,在这种场景下,用户对近期吃过的东西一般没兴趣,但随着时间的推移,用户会对之前吃过的东西,逐渐变得有新鲜感,并且距离之前吃该东西的时间越久,该新鲜感会越强,因此需要适时向用户推荐之前消费过的商品,这就是时间因素对推荐过程的正向影响随时间增长的表现。
基于上述,以第一用户-商户初始评价矩阵中的用户U a对在商户S k处消费过的n个商品做出评价为例,n个商品可表示为<i 1,i 2,...,i n>,则第一用户-商户初始评价矩阵中商户S k对应的用户U a的商品评价值序列可表示为
Figure PCTCN2017118777-appb-000035
该商品评价值序列对应的网络行为发生时间序列可表示为
Figure PCTCN2017118777-appb-000036
即,用户U a在时刻
Figure PCTCN2017118777-appb-000037
在商户S k处消费商品i 1,并对商品i 1形成评价值
Figure PCTCN2017118777-appb-000038
相应地,用户U a在时刻
Figure PCTCN2017118777-appb-000039
在商户S k处消费商品i 2,并对商品i 2形成评价值
Figure PCTCN2017118777-appb-000040
……,依次类推,用户U a在时刻
Figure PCTCN2017118777-appb-000041
在商户S k处消费商品i n,并对商品i n形成评价值
Figure PCTCN2017118777-appb-000042
基于上述,可以采用下述公式(1),根据推荐场景下的时间影响特性以及网络行为发生时间序列
Figure PCTCN2017118777-appb-000043
计算商品评价值序列
Figure PCTCN2017118777-appb-000044
Figure PCTCN2017118777-appb-000045
对应的时间修正因子序列
Figure PCTCN2017118777-appb-000046
中的时间修正因子。
Figure PCTCN2017118777-appb-000047
在公式(1)中,
Figure PCTCN2017118777-appb-000048
表示商品评价值序列
Figure PCTCN2017118777-appb-000049
中评价值
Figure PCTCN2017118777-appb-000050
对 应的时间修正因子,1≤j≤n,j和n是自然数。T now表示当前时间。T period表示推荐场景下的时间周期,为一预设值或经验值,例如在外卖场景中,该时间周期的取值可以是5、7、10或14等。
Figure PCTCN2017118777-appb-000051
表示评价值
Figure PCTCN2017118777-appb-000052
对应的网络行为发生时间。
进一步,基于上述公式(1)计算的时间修正因子,可以采用下述公式(2),将所述商品评价值序列
Figure PCTCN2017118777-appb-000053
与其对应的时间修正因子序列
Figure PCTCN2017118777-appb-000054
相乘,以获得用户-商户修正评价矩阵中商户S k对应的用户U a的修正评价值。其中,商户S k对应的用户U a的修正评价值也就是用户U a对商户S k的修正评价值。
Figure PCTCN2017118777-appb-000055
在公式(2)中,v' ak表示商户S k对应的用户U a的修正评价值,其它参数的解释参见公式(1),在此不再赘述。
基于上述过程,可以根据推荐场景下的时间影响特性,对第一用户-商户初始评价矩阵中的各元素值进行修正,从而得到用户-商户修正评价值矩阵。
在本实施例中,线下构建用户-商户修正评价矩阵,使得在线上进行推荐时,可以直接从中获取候选商户对应的至少一个用户的修正评价值,有利于提高线上推荐的效率。
当然,除了线下预先构建用户-商户修正评价矩阵,在线上进行推荐时,直接从中获取候选商户对应的至少一个用户的修正评价值之外,也可以在线上进行推荐时,实时计算候选商户对应的至少一个用户的修正评价值,然后基于此向待推荐用户推荐商户。
图4为本公开又一实施例提供的推荐方法的流程示意图。如图4所示,所述方法包括:
401、根据待推荐用户,确定至少一个候选商户。
402、基于系统中存在的用户针对至少一个候选商户提供的商品的网络行为信息,构建第二用户-商品评价矩阵。
403、对第二用户-商品评价矩阵进行维度切换,以获得第二用户-商户初始评价矩阵,所述第二用户-商户初始评价矩阵包括至少一个候选商户各自对应的至少一个用户的商品评价值序列。
404、根据推荐场景下的时间影响特性对第二用户-商户初始评价矩阵进行修正,以获得用户-候选商户修正评价矩阵,所述用户-候选商户修正评价矩阵包括至少一个候选商户各自对应的至少一个用户的修正评价值。
405、基于至少一个候选商户各自对应的至少一个用户的修正评价值,向所述待推荐用户推荐商户。
关于步骤401可参见前述实施例中相应步骤的描述,在此不再赘述。
步骤402-404,用于在线实时生成至少一个候选商户各自对应的至少一个用户的修正评价值的过程,其过程类似图3a所示预先构建用户-商户修正评价矩阵的过程,区别仅在于:数据集不相同。关于步骤402-404的详细过程可参考图3a所示实施例的过程,在此不再详述。
在获得至少一个候选商户各自对应的至少一个用户的修正评价值之后,如步骤405所述,可基于至少一个候选商户各自对应的至少一个用户的修正评价值,向待推荐用户推荐商户,提高推荐效果。
在上述实施例或下述实施例中,可基于至少一个候选商户各自对应的至少一个用户的修正评价值,向待推荐用户推荐商户。
可选地,可以采用基于用户的协同过滤算法,基于至少一个候选商户各自对应的至少一个用户的修正评价值,向待推荐用户推荐商户。或者
可选地,可以采用基于商户的协同过滤算法,基于至少一个候选商户各自对应的至少一个用户的修正评价值,向待推荐用户推荐商户。
下面通过不同的实施例分别对上述基于用户的协同过滤算法的推荐过程和上述基于商户的协同过滤算法的推荐过程进行详细说明。
图5为本公开又一实施例提供的推荐方法的流程示意图。如图5所 示,所述方法包括:
501、根据待推荐用户的位置,从商户集合中,确定至少一个候选商户。
502、获取至少一个候选商户各自对应的至少一个用户的修正评价值,所述修正评价值是按照推荐场景下的时间影响特性对初始评价值修正得到的。
在本实施例中,步骤502的具体实现可以参照图2a和图3a所示实施方式,或者也可以参照图4所示实施例中的步骤402-404来实现,在此均不做赘述。
503、从第一用户集合中,获取待推荐用户的相似用户。
可选地,所述第一用户集合可以由系统中存在的所有用户组成,或者也可以由系统中存在的部分用户组成,或者还可以由至少一个候选商户各自对应的至少一个用户组成。换句话说,一种获取方式为:从系统中存在的所有用户中,获取待推荐用户的相似用户。另一种获取方式为:从系统中存在的部分用户中,获取待推荐用户的相似用户。又一种获取方式为:从至少一个候选商户各自对应的至少一个用户中,获取待推荐用户的相似用户。
可选地,可以采用余弦相似性计算方法或皮尔森(Pearson)相似性计算方法,从第一用户集合中,获取待推荐用户的相似用户。
优选地,可以采用Pearson相似性计算方法,计算待推荐用户与第一用户集合中各用户的相似性,Pearson相似性计算方法具有相似度计算精度较高的优势。将待推荐用户表示为u a,将第一用户集合中任一用户表示为u b,则可以采用如下公式(3)计算u a与u b的Pearson相似性。
Figure PCTCN2017118777-appb-000056
在公式(3)中,M表示u a和u b的相交商户的个数,所述相交商户 是指u a和u b存在交集的(例如两者都评价过的)商户;v a,j表示u a对相交商户中的商户s j的评价值;
Figure PCTCN2017118777-appb-000057
表示u a的平均评价值,v b,j表示u b对商户s j的评价值,
Figure PCTCN2017118777-appb-000058
表示u b的平均评价值,w a,b表示u a与u b的Pearson相似性。
在上述公式(3)中,实际上只根据u a和u b的相交商户计算了u a和u b的相似性。实际应用中,u a和u b的相交商户的数量可能很少,为此可以采用下述公式(4)对上述公式(3)计算出的相似性进行修正,获得修正后的相似性。
Figure PCTCN2017118777-appb-000059
在公式(4)中,w′ a,b表示u a和u b的修正后的相似性。F表示一个常数,其取值可视具体应用场景而定,例如可以是但不限于:10。其中,将
Figure PCTCN2017118777-appb-000060
称为影响因子,且
Figure PCTCN2017118777-appb-000061
即当u a和u b的相交商户的数量M大于或等于常数F时,说明两者相交商户的数量较多,无需用影响因子进行修正,故影响因子取值为1;反之,当u a和u b的相交商户的数量M小于常数F时,说明两者相交商户的数量较少,需用影响因子进行修正,且影响因子的取值小于1。采用这个影响因子能提高相似性计算结果的准确性和精度。
由此可见,基于上述公式(3)和(4),可以计算出待推荐用户与第一用户集合中每个用户的相似性,进而可基于待推荐用户与第一用户集合中每个用户的相似性,从第一用户集合中确定待推荐用户的相似用户。例如,可以选择与待推荐用户的相似性最高的至少一个用户作为待推荐用户的相似用户。又例如,可以选择与待推荐用户的相似性大于设定阈值的至少一个用户作为待推荐用户的相似用户。
504、根据待推荐用户的相似用户对至少一个候选商户的修正评价 值,计算待推荐用户对至少一个候选商户的修正评价值。
可选地,可以根据下述公式(5),计算待推荐用户对至少一个候选商户中每个候选商户的修正评价值。
Figure PCTCN2017118777-appb-000062
其中,p a,i表示待推荐用户u a对至少一个候选商户中候选商户s i的修正评价值,s i∈S,S表示至少一个候选商户组成的候选商户集合。U'表示由待推荐用户的相似用户构成的相似用户集合;w′ a,c表示待推荐用户u a与相似用户集合中的相似用户u c的修正后的相似性;v c,i表示相似用户集合中的相似用户u c对至少一个候选商户中候选商户s i的修正评价值;
Figure PCTCN2017118777-appb-000063
表示相似用户集合中的相似用户u c的平均评价值。
举例说明:假设,用户对候选商户的修正评价值的范围为0到5,依次表示不喜欢到喜欢的程度,假设需要根据下述表1所示用户User1-User5对候选商户Supplier_1-Supplier_6的修正评价值,计算待推荐用户User5对候选商户Supplier_5的修正评价值。
从表1可以看出,与User5最相似的用户是User3、然后是User4、User2和User1,利用这些相似用户对Supplier_5的修正评价值,可以计算User5对Supplier_5的修正评价值。其中,从Supplier_5这一列可以发现,相似用户对Supplier_5的评价都不错,因此,User5对Supplier_5的评价也会不错。
表1
Figure PCTCN2017118777-appb-000064
其中,可以采用上述公式(3)和(4)发现User5的相似用户;根据上述公式(5)计算User5对Supplier_5的修正评价值。
505、根据待推荐用户对至少一个候选商户的修正评价值,从至少一个候选商户中向待推荐用户推荐商户。
可选地,可以选择修正评价值最大的至少一个候选商户推荐给待推荐用户。或者,可以选择修正评价值大于设定阈值的至少一个候选商户推荐给待推荐用户。
在本实施中,考虑时间因素的影响,基于符合推荐场景的时间影响特性,修正用户对商品的初始评价值,进而基于修正后的评价值进行推荐,达到推荐过程考虑时间因素的目的,从而提高推荐结果的准确性。另外,在本实施例中,基于用户的协同过滤算法利用用户之间的相似性进行商户推荐,具有很好的预测精度,而且能随数据的变化而变化,比较适合于数据更新频繁的系统。
图6为本公开又一实施例提供的推荐方法的流程示意图。如图6所示,所述方法包括:
601、根据待推荐用户的位置,从商户集合中,确定至少一个候选商户。
602、获取至少一个候选商户各自对应的至少一个用户的修正评价值,所述修正评价值是按照推荐场景下的时间影响特性对初始评价值修正得到的。
603、从第一商户集合中,获取至少一个候选商户各自的相似商户。
可选地,可以选择修正评价值最大的至少一个候选商户推荐给待推荐用户。或者,可以选择修正评价值大于设定阈值的至少一个候选商户推荐给待推荐用户。
可选地,第一商户集合可以由至少一个商户组成,或者可以由系统中存在的所有商户组成,或者可以由系统中存在的部分商户组成。换句话说,一种获取方式为:在至少一个候选商户中,获取至少一个候选商户各自的相似商户。另一种获取方式为:在系统中存在的所有商户中,获取至少一个候选商户各自的相似商户。又一种获取方式为:在系统中 存在的部分商户中,获取至少一个候选商户各自的相似商户。
可选地,可以采用余弦相似性计算方法或Pearson相似性计算方法,从第一商户集合中,获取每个候选商户的相似商户。
值得说明的是,采用余弦相似性计算方法或Pearson相似性计算方法,计算各候选商户与第一商户集合中各商户的相似性的过程,类似根据上述公式(3)和公式(4)计算用户间相似性的过程,区别仅在于:需要将公式(3)和公式(4)中的用户相关参数替换为候选商户和第一商户集合中各商户的相关参数,具体计算过程在此不再详述。
604、根据至少一个候选商户各自的相似商户对应的用户的修正评价值,计算待推荐用户对至少一个候选商户的修正评价值。
605、根据待推荐用户对至少一个候选商户的修正评价值,从至少一个候选商户中向待推荐用户推荐商户。
在本实施中,考虑时间因素的影响,基于符合推荐场景的时间影响特性,修正用户对商品的初始评价值,进而基于修正后的评价值进行推荐,达到推荐过程考虑时间因素的目的,从而提高推荐结果的准确性。另外,在本实施例中,基于商户的协同过滤算法利用商户之间的相似性进行商户推荐,可以解决数据稀疏的问题,有利于提高推荐精度。
需要说明的是,上述实施例所提供方法的各步骤的执行主体均可以是同一设备,或者,该方法也由不同设备作为执行主体。比如,步骤101至步骤103的执行主体可以为设备A;又比如,步骤101和102的执行主体可以为设备A,步骤103的执行主体可以为设备B;等等。
图7为本公开又一实施例提供的推荐装置的结构示意图。如图7所示,所述推荐装置包括:确定模块71、获取模块72以及推荐模块73。
确定模块71,被配置为根据待推荐用户,确定至少一个候选商户。
获取模块72,被配置为获取至少一个候选商户各自对应的至少一个用户的修正评价值,所述修正评价值是按照推荐场景下的时间影响特性对初始评价值修正得到的。
推荐模块73,被配置为基于至少一个候选商户各自对应的至少一个用户的修正评价值,向待推荐用户推荐商户。
在一可选实施方式中,获取模块72具体被配置为:从预先构建的用户-商户修正评价矩阵中,获取至少一个候选商户各自对应的至少一个用户的修正评价值;其中,所述用户-商户修正评价矩阵包括系统中多个商户各自对应的至少一个用户的修正评价值,所述多个商户包含所述至少一个候选商户。
在一可选实施方式中,如图8所示,推荐装置还包括:构建模块74,被配置为预先构建用户-商户修正评价矩阵。
如图8所示,构建模块74的一种实现结构包括:构建子模块741、维度切换子模块742以及修正子模块743。
构建子模块741,被配置为基于系统中存在的用户针对商品的网络行为信息,构建第一用户-商品评价矩阵。
维度切换子模块742,被配置为对第一用户-商品评价矩阵进行维度切换,以获得第一用户-商户初始评价矩阵,所述第一用户-商户初始评价矩阵包括所述多个商户各自对应的至少一个用户的商品评价值序列。
修正子模块743,被配置为根据推荐场景下的时间影响特性对第一用户-商户初始评价矩阵进行修正,以获得用户-商户修正评价矩阵。
在一可选实施方式中,修正子模块743具体被配置为:
根据推荐场景下的时间影响特性以及第一用户-商户初始评价矩阵中各商品评价值序列对应的网络行为发生时间序列,计算第一用户-商户初始评价矩阵中各商品评价值序列对应的时间修正因子序列;
将第一用户-商户初始评价矩阵中各商品评价值序列与各自对应的时间修正因子序列相乘,以获得用户-商户修正评价矩阵。
在一可选实施方式中,所述推荐场景下的时间影响特性为:时间因素对推荐过程的正向影响随时间增长的特性。
在一可选实施方式中,对第一用户-商户初始评价矩阵中商户S k对应的用户U a的商品评价值序列
Figure PCTCN2017118777-appb-000065
其对应的网络行为发生时间序列为
Figure PCTCN2017118777-appb-000066
则修正子模块743具体可以根据上述公式(1)计算商品评价值序列
Figure PCTCN2017118777-appb-000067
对应的时间修正因子序列
Figure PCTCN2017118777-appb-000068
Figure PCTCN2017118777-appb-000069
在一可选实施方式中,修正子模块743具体可以根据上述公式(2)将商品评价值序列
Figure PCTCN2017118777-appb-000070
与其对应的时间修正因子序列
Figure PCTCN2017118777-appb-000071
Figure PCTCN2017118777-appb-000072
相乘,以获得用户-商户修正评价矩阵中商户S k对应的用户U a的修正评价值。
关于公式(1)与公式(2)以及其中有关参数的描述,可参见前述方法实施例,在此不再赘述。
在一可选实施方式中,获取模块72具体被配置为:基于系统中存在的用户针对至少一个候选商户提供的商品的网络行为信息,构建第二用户-商品评价矩阵;对第二用户-商品评价矩阵进行维度切换,以获得第二用户-商户初始评价矩阵,所述第二用户-商户初始评价矩阵包括至少一个候选商户各自对应的至少一个用户的商品评价值序列;根据推荐场景下的时间影响特性对第二用户-商户初始评价矩阵进行修正,以获得用户-候选商户修正评价矩阵,所述用户-候选商户修正评价矩阵包括所述至少一个候选商户各自对应的至少一个用户的修正评价值。
在一可选实施方式中,推荐模块73具体被配置为:采用基于用户的协同过滤算法,基于所述至少一个候选商户各自对应的至少一个用户的修正评价值,向所述待推荐用户推荐商户;或者,采用基于商户的协同过滤算法,基于所述至少一个候选商户各自对应的至少一个用户的修正评价值,向所述待推荐用户推荐商户。
在一可选实施方式中,确定模块71具体被配置为:根据待推荐用户的位置,从商户集合中,获取至少一个候选商户。
本公开实施例提供的推荐装置,可用于执行上述方法实施例提供的流程,其具体工作原理不再赘述。
本实施例提供的推荐装置,在推荐过程中,考虑时间因素的影响,基于符合推荐场景的时间影响特性,修正用户对商品的初始评价值,进而基于修正后的评价值进行推荐,达到推荐过程考虑时间因素的目的, 从而提高推荐结果的准确性。
本领域内的技术人员应明白,本公开的实施例可提供为方法、系统、或计算机程序产品。因此,本公开可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且,本公开可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本公开是参照根据本公开实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器,使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中,使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品,该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上,使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理,从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
在一个典型的配置中,计算设备包括一个或多个处理器(CPU)、输入/输出接口、网络接口和内存。
内存可能包括计算机可读介质中的非永久性存储器,随机存取存储器(RAM)和/或非易失性内存等形式,如只读存储器(ROM)或闪存 (flash RAM)。内存是计算机可读介质的示例。
计算机可读介质包括永久性和非永久性、可移动和非可移动媒体可以由任何方法或技术来实现信息存储。信息可以是计算机可读指令、数据结构、程序的模块或其他数据。计算机的存储介质的例子包括,但不限于相变内存(PRAM)、静态随机存取存储器(SRAM)、动态随机存取存储器(DRAM)、其他类型的随机存取存储器(RAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、快闪记忆体或其他内存技术、只读光盘只读存储器(CD-ROM)、数字多功能光盘(DVD)或其他光学存储、磁盒式磁带,磁带磁磁盘存储或其他磁性存储设备或任何其他非传输介质,可用于存储可以被计算设备访问的信息。按照本文中的界定,计算机可读介质不包括暂存电脑可读媒体(transitory media),如调制的数据信号和载波。
还需要说明的是,术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、方法、商品或者设备不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、方法、商品或者设备所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括所述要素的过程、方法、商品或者设备中还存在另外的相同要素。
本领域技术人员应明白,本公开的实施例可提供为方法、系统或计算机程序产品。因此,本公开可采用完全硬件实施例、完全软件实施例或结合软件和硬件方面的实施例的形式。而且,本公开可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
以上所述仅为本公开的实施例而已,并不用于限制本公开。对于本领域技术人员来说,本公开可以有各种更改和变化。凡在本公开的精神和原理之内所作的任何修改、等同替换、改进等,均应包含在本公开的权利要求范围之内。

Claims (22)

  1. 一种推荐方法,包括:
    根据待推荐用户,确定至少一个候选商户;
    获取所述至少一个候选商户各自对应的至少一个用户的修正评价值,所述修正评价值是按照推荐场景下的时间影响特性对初始评价值修正得到的;
    基于所述至少一个候选商户各自对应的至少一个用户的修正评价值,向所述待推荐用户推荐商户。
  2. 根据权利要求1所述的方法,其中,所述获取所述至少一个候选商户各自对应的至少一个用户的修正评价值,包括:
    从预先构建的用户-商户修正评价矩阵中,获取所述至少一个候选商户各自对应的至少一个用户的修正评价值;
    其中,所述用户-商户修正评价矩阵包括系统中多个商户各自对应的至少一个用户的修正评价值,所述多个商户包含所述至少一个候选商户。
  3. 根据权利要求2所述的方法,其中,预先构建用户-商户修正评价矩阵,包括:
    基于系统中存在的用户针对商品的网络行为信息,构建第一用户-商品评价矩阵;
    对所述第一用户-商品评价矩阵进行维度切换,以获得第一用户-商户初始评价矩阵,所述第一用户-商户初始评价矩阵包括所述多个商户各自对应的至少一个用户的商品评价值序列;
    根据所述推荐场景下的时间影响特性对所述第一用户-商户初始评价矩阵进行修正,以获得所述用户-商户修正评价矩阵。
  4. 根据权利要求3所述的方法,其中,所述根据所述推荐场景下的时间影响特性对所述第一用户-商户初始评价矩阵进行修正,以获得所述用户-商户修正评价矩阵,包括:
    根据所述推荐场景下的时间影响特性以及所述第一用户-商户初始评价矩阵中各商品评价值序列对应的网络行为发生时间序列,计算所述 第一用户-商户初始评价矩阵中各商品评价值序列对应的时间修正因子序列;
    将所述第一用户-商户初始评价矩阵中各商品评价值序列与各自对应的时间修正因子序列相乘,以获得所述用户-商户修正评价矩阵。
  5. 根据权利要求4所述的方法,其中,所述推荐场景下的时间影响特性为:时间因素对推荐过程的正向影响随时间增长的特性。
  6. 根据权利要求5所述的方法,其中,对所述第一用户-商户初始评价矩阵中商户S k对应的用户U a的商品评价值序列
    Figure PCTCN2017118777-appb-100001
    其对应的网络行为发生时间序列为
    Figure PCTCN2017118777-appb-100002
    所述根据所述推荐场景下的时间影响特性以及所述网络行为发生时间序列
    Figure PCTCN2017118777-appb-100003
    计算所述商品评价值序列
    Figure PCTCN2017118777-appb-100004
    对应的时间修正因子序列
    Figure PCTCN2017118777-appb-100005
    包括:
    根据公式
    Figure PCTCN2017118777-appb-100006
    计算所述商品评价值序列
    Figure PCTCN2017118777-appb-100007
    对应的时间修正因子序列
    Figure PCTCN2017118777-appb-100008
    中的时间修正因子;
    Figure PCTCN2017118777-appb-100009
    表示所述商品评价值序列
    Figure PCTCN2017118777-appb-100010
    中评价值
    Figure PCTCN2017118777-appb-100011
    对应的时间修正因子,1≤j≤n,j和n是自然数;
    T now表示当前时间;
    T period表示所述推荐场景下的时间周期;
    Figure PCTCN2017118777-appb-100012
    表示评价值
    Figure PCTCN2017118777-appb-100013
    对应的网络行为发生时间。
  7. 根据权利要求6所述的方法,其中,将所述商品评价值序列
    Figure PCTCN2017118777-appb-100014
    Figure PCTCN2017118777-appb-100015
    与其对应的时间修正因子序列
    Figure PCTCN2017118777-appb-100016
    相乘,以获得所述用户-商户修正评价矩阵中商户S k对应的用户U a的修正评价值,包括:
    根据公式
    Figure PCTCN2017118777-appb-100017
    计算所述商户S k对应的用户U a的修正评价值;
    v′ ak表示所述商户S k对应的用户U a的修正评价值。
  8. 根据权利要求1所述的方法,其中,所述获取所述至少一个候选商户各自对应的至少一个用户的修正评价值,包括:
    基于系统中存在的用户针对所述至少一个候选商户提供的商品的网络行为信息,构建第二用户-商品评价矩阵;
    对所述第二用户-商品评价矩阵进行维度切换,以获得第二用户-商户初始评价矩阵,所述第二用户-商户初始评价矩阵包括所述至少一个候选商户各自对应的至少一个用户的商品评价值序列;
    根据所述推荐场景下的时间影响特性对所述第二用户-商户初始评价矩阵进行修正,以获得用户-候选商户修正评价矩阵,所述用户-候选商户修正评价矩阵包括所述至少一个候选商户各自对应的至少一个用户的修正评价值。
  9. 根据权利要求1-8任一项所述的方法,其中,所述基于所述至少一个候选商户各自对应的至少一个用户的修正评价值,向所述待推荐用户推荐商户,包括:
    采用基于用户的协同过滤算法,基于所述至少一个候选商户各自对应的至少一个用户的修正评价值,向所述待推荐用户推荐商户;或者
    采用基于商户的协同过滤算法,基于所述至少一个候选商户各自对应的至少一个用户的修正评价值,向所述待推荐用户推荐商户。
  10. 根据权利要求1-8任一项所述的方法,其中,所述根据待推荐用户,确定至少一个候选商户,包括:
    根据所述待推荐用户的位置,从商户集合中,获取所述至少一个候选商户。
  11. 一种推荐装置,包括:
    确定模块,被配置为根据待推荐用户,确定至少一个候选商户;
    获取模块,被配置为获取所述至少一个候选商户各自对应的至少一个用户的修正评价值,所述修正评价值是按照推荐场景下的时间影响特性对初始评价值修正得到的;
    推荐模块,被配置为基于所述至少一个候选商户各自对应的至少一个用户的修正评价值,向所述待推荐用户推荐商户。
  12. 根据权利要求11所述的装置,其中,所述获取模块具体被配置为:
    从预先构建的用户-商户修正评价矩阵中,获取所述至少一个候选商户各自对应的至少一个用户的修正评价值;
    其中,所述用户-商户修正评价矩阵包括系统中多个商户各自对应的至少一个用户的修正评价值,所述多个商户包含所述至少一个候选商户。
  13. 根据权利要求12所述的装置,还包括:
    构建模块,被配置为预先构建所述用户-商户修正评价矩阵;
    所述构建模块包括:
    构建子模块,被配置为基于系统中存在的用户针对商品的网络行为信息,构建第一用户-商品评价矩阵;
    维度切换子模块,被配置为对所述第一用户-商品评价矩阵进行维度切换,以获得第一用户-商户初始评价矩阵,所述第一用户-商户初始评价矩阵包括所述多个商户各自对应的至少一个用户的商品评价值序列;
    修正子模块,被配置为根据所述推荐场景下的时间影响特性对所述第一用户-商户初始评价矩阵进行修正,以获得所述用户-商户修正评价矩阵。
  14. 根据权利要求13所述的装置,其中,所述修正子模块具体被配置为:
    根据所述推荐场景下的时间影响特性以及所述第一用户-商户初始评价矩阵中各商品评价值序列对应的网络行为发生时间序列,计算所述第一用户-商户初始评价矩阵中各商品评价值序列对应的时间修正因子序列;
    将所述第一用户-商户初始评价矩阵中各商品评价值序列与各自对 应的时间修正因子序列相乘,以获得所述用户-商户修正评价矩阵。
  15. 根据权利要求14所述的装置,其中,所述推荐场景下的时间影响特性为:时间因素对推荐过程的正向影响随时间增长的特性。
  16. 根据权利要求15所述的装置,其中,对所述第一用户-商户初始评价矩阵中商户S k对应的用户U a的商品评价值序列
    Figure PCTCN2017118777-appb-100018
    其对应的网络行为发生时间序列为
    Figure PCTCN2017118777-appb-100019
    所述修正子模块在根据所述推荐场景下的时间影响特性以及所述网络行为发生时间序列
    Figure PCTCN2017118777-appb-100020
    Figure PCTCN2017118777-appb-100021
    计算所述商品评价值序列
    Figure PCTCN2017118777-appb-100022
    对应的时间修正因子序列
    Figure PCTCN2017118777-appb-100023
    时,具体被配置为:
    根据公式
    Figure PCTCN2017118777-appb-100024
    计算所述商品评价值序列
    Figure PCTCN2017118777-appb-100025
    对应的时间修正因子序列
    Figure PCTCN2017118777-appb-100026
    中的时间修正因子;
    Figure PCTCN2017118777-appb-100027
    表示所述商品评价值序列
    Figure PCTCN2017118777-appb-100028
    中评价值
    Figure PCTCN2017118777-appb-100029
    对应的时间修正因子,1≤j≤n,j和n是自然数;
    T now表示当前时间;
    T period表示所述推荐场景下的时间周期;
    Figure PCTCN2017118777-appb-100030
    表示评价值
    Figure PCTCN2017118777-appb-100031
    对应的网络行为发生时间。
  17. 根据权利要求16所述的装置,其中,所述修正子模块在将所述商品评价值序列
    Figure PCTCN2017118777-appb-100032
    与其对应的时间修正因子序列
    Figure PCTCN2017118777-appb-100033
    Figure PCTCN2017118777-appb-100034
    相乘,以获得所述用户-商户修正评价矩阵中商户S k对应的用户U a的修正评价值时,具体被配置为:
    根据公式
    Figure PCTCN2017118777-appb-100035
    计算所述商户S k对应的用户U a的修正评价 值;
    v′ ak表示所述商户S k对应的用户U a的修正评价值。
  18. 根据权利要求11所述的装置,其中,所述获取模块具体被配置为:
    基于系统中存在的用户针对所述至少一个候选商户提供的商品的网络行为信息,构建第二用户-商品评价矩阵;
    对所述第二用户-商品评价矩阵进行维度切换,以获得第二用户-商户初始评价矩阵,所述第二用户-商户初始评价矩阵包括所述至少一个候选商户各自对应的至少一个用户的商品评价值序列;
    根据所述推荐场景下的时间影响特性对所述第二用户-商户初始评价矩阵进行修正,以获得用户-候选商户修正评价矩阵,所述用户-候选商户修正评价矩阵包括所述至少一个候选商户各自对应的至少一个用户的修正评价值。
  19. 根据权利要求11-18任一项所述的装置,其中,所述推荐模块具体被配置为:
    采用基于用户的协同过滤算法,基于所述至少一个候选商户各自对应的至少一个用户的修正评价值,向所述待推荐用户推荐商户;或者
    采用基于商户的协同过滤算法,基于所述至少一个候选商户各自对应的至少一个用户的修正评价值,向所述待推荐用户推荐商户。
  20. 根据权利要求11-18任一项所述的装置,其中,所述确定模块具体被配置为:
    根据所述待推荐用户的位置,从商户集合中,获取所述至少一个候选商户。
  21. 一种电子设备,包括:存储器和处理器;所述存储器用于存储一条或多条计算机指令,其中,所述一条或多条计算机指令被所述处理器执行时能够实现权利要求1-10任一项所述方法中的步骤。
  22. 一种存储有计算机程序的计算机可读存储介质,其中,所述计算机程序使计算机执行权利要求1-10任一项所述方法中的步骤。
PCT/CN2017/118777 2017-04-26 2017-12-26 推荐方法及装置 WO2018196424A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/663,305 US20200126035A1 (en) 2017-04-26 2019-10-24 Recommendation method and apparatus, electronic device, and computer storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710282964.3 2017-04-26
CN201710282964.3A CN107169052B (zh) 2017-04-26 2017-04-26 推荐方法及装置

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US16/663,305 Continuation US20200126035A1 (en) 2017-04-26 2019-10-24 Recommendation method and apparatus, electronic device, and computer storage medium

Publications (1)

Publication Number Publication Date
WO2018196424A1 true WO2018196424A1 (zh) 2018-11-01

Family

ID=59812887

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/118777 WO2018196424A1 (zh) 2017-04-26 2017-12-26 推荐方法及装置

Country Status (3)

Country Link
US (1) US20200126035A1 (zh)
CN (1) CN107169052B (zh)
WO (1) WO2018196424A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111128376A (zh) * 2019-11-21 2020-05-08 泰康保险集团股份有限公司 一种推荐评估表单的方法和装置
CN113077312A (zh) * 2021-04-12 2021-07-06 上海华客信息科技有限公司 酒店推荐方法、系统、设备及存储介质
CN113361920A (zh) * 2021-06-04 2021-09-07 上海华客信息科技有限公司 酒店服务优化指标推荐方法、系统、设备及存储介质
CN113537506A (zh) * 2020-04-22 2021-10-22 百度在线网络技术(北京)有限公司 针对机器学习效果的测试方法、装置、设备和介质
US20230094255A1 (en) * 2021-09-27 2023-03-30 7-Eleven, Inc. Autonomous delivery mechanism data integration in an application platform

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106530010B (zh) * 2016-11-15 2017-12-12 平安科技(深圳)有限公司 融合时间因素的协同过滤方法和装置
CN107169052B (zh) * 2017-04-26 2019-03-05 北京小度信息科技有限公司 推荐方法及装置
CN107742245B (zh) * 2017-10-31 2021-04-27 北京星选科技有限公司 一种商户信息推荐方法、装置及设备
CN108154425B (zh) * 2018-01-19 2021-03-26 广州天源信息科技股份有限公司 一种结合社会网络和位置的线下商户推荐方法
CN110264290B (zh) * 2018-03-12 2021-07-30 北京三快在线科技有限公司 获取推荐信息的方法、装置及电子设备
CN111563206B (zh) * 2020-07-03 2024-01-12 秦皇岛肆拾贰宇数字科技有限公司 物品信息序列显示方法、装置、电子设备和介质
CN112507209B (zh) * 2020-11-10 2022-07-05 中国科学院深圳先进技术研究院 一种基于陆地移动距离进行知识蒸馏的序列推荐方法
CN112465564A (zh) * 2020-12-11 2021-03-09 广东电网有限责任公司 一种供应商推荐方法、装置、终端
CN115080833B (zh) * 2021-03-10 2024-02-27 阿里巴巴新加坡控股有限公司 一种信息流推荐方法、设备、系统及存储介质
CN113283960B (zh) * 2021-05-18 2022-08-19 河南中钢网科技集团股份有限公司 基于大数据分析和云计算的垂直电商平台商品智能推荐方法和云服务平台

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110213661A1 (en) * 2010-03-01 2011-09-01 Joseph Milana Computer-Implemented Method For Enhancing Product Sales
CN105096152A (zh) * 2014-05-20 2015-11-25 阿里巴巴集团控股有限公司 一种基于商品热度的操作执行方法及装置
CN105469263A (zh) * 2014-09-24 2016-04-06 阿里巴巴集团控股有限公司 一种商品推荐方法及装置
CN105512914A (zh) * 2015-12-09 2016-04-20 联想(北京)有限公司 一种信息处理方法及电子设备
CN105677831A (zh) * 2016-01-04 2016-06-15 拉扎斯网络科技(上海)有限公司 一种确定推荐商户的方法及装置
CN106228386A (zh) * 2016-07-12 2016-12-14 腾讯科技(深圳)有限公司 一种信息推送方法及装置
CN107169052A (zh) * 2017-04-26 2017-09-15 北京小度信息科技有限公司 推荐方法及装置

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104239335B (zh) * 2013-06-19 2017-10-27 阿里巴巴集团控股有限公司 特定用户信息获取方法及装置
CN104899462B (zh) * 2015-06-17 2017-09-29 北京奇虎科技有限公司 确定信息发布位置的价值度的方法及装置

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110213661A1 (en) * 2010-03-01 2011-09-01 Joseph Milana Computer-Implemented Method For Enhancing Product Sales
CN105096152A (zh) * 2014-05-20 2015-11-25 阿里巴巴集团控股有限公司 一种基于商品热度的操作执行方法及装置
CN105469263A (zh) * 2014-09-24 2016-04-06 阿里巴巴集团控股有限公司 一种商品推荐方法及装置
CN105512914A (zh) * 2015-12-09 2016-04-20 联想(北京)有限公司 一种信息处理方法及电子设备
CN105677831A (zh) * 2016-01-04 2016-06-15 拉扎斯网络科技(上海)有限公司 一种确定推荐商户的方法及装置
CN106228386A (zh) * 2016-07-12 2016-12-14 腾讯科技(深圳)有限公司 一种信息推送方法及装置
CN107169052A (zh) * 2017-04-26 2017-09-15 北京小度信息科技有限公司 推荐方法及装置

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111128376A (zh) * 2019-11-21 2020-05-08 泰康保险集团股份有限公司 一种推荐评估表单的方法和装置
CN111128376B (zh) * 2019-11-21 2023-06-16 泰康保险集团股份有限公司 一种推荐评估表单的方法和装置
CN113537506A (zh) * 2020-04-22 2021-10-22 百度在线网络技术(北京)有限公司 针对机器学习效果的测试方法、装置、设备和介质
CN113537506B (zh) * 2020-04-22 2023-08-29 百度在线网络技术(北京)有限公司 针对机器学习效果的测试方法、装置、设备和介质
CN113077312A (zh) * 2021-04-12 2021-07-06 上海华客信息科技有限公司 酒店推荐方法、系统、设备及存储介质
CN113361920A (zh) * 2021-06-04 2021-09-07 上海华客信息科技有限公司 酒店服务优化指标推荐方法、系统、设备及存储介质
US20230094255A1 (en) * 2021-09-27 2023-03-30 7-Eleven, Inc. Autonomous delivery mechanism data integration in an application platform

Also Published As

Publication number Publication date
CN107169052B (zh) 2019-03-05
CN107169052A (zh) 2017-09-15
US20200126035A1 (en) 2020-04-23

Similar Documents

Publication Publication Date Title
WO2018196424A1 (zh) 推荐方法及装置
CN110162693B (zh) 一种信息推荐的方法以及服务器
WO2019165872A1 (zh) 营销产品的推荐方法和装置
JP5897019B2 (ja) 候補製品のリンクリストを判定する方法および装置
US9021429B1 (en) Recommending applications to portable electronic devices
US10354184B1 (en) Joint modeling of user behavior
US20170206276A1 (en) Large Scale Recommendation Engine Based on User Tastes
WO2018014771A1 (zh) 一种数据对象推送的方法、装置及系统
US20160283481A1 (en) Method and apparatus for combining text search and recommendation engines
WO2020233432A1 (zh) 一种信息推荐方法及装置
US20180308152A1 (en) Data Processing Method and Apparatus
US10937070B2 (en) Collaborative filtering to generate recommendations
WO2020221022A1 (zh) 业务对象推荐方法
US20220172083A1 (en) Noise contrastive estimation for collaborative filtering
US20210383259A1 (en) Dynamic workflow optimization using machine learning techniques
Singh et al. A survey on the generation of recommender systems
Xin et al. ATNN: adversarial two-tower neural network for new item’s popularity prediction in E-commerce
CN112100511B (zh) 一种偏好程度数据获得方法、装置以及电子设备
CN111737537B (zh) 基于图数据库的poi推荐方法、设备及介质
Zhao A Study on the Application of Big Data Mining in E-commerce
US10380205B2 (en) Algorithm for selecting and scoring suggested action
CN116861107A (zh) 业务内容展示方法、装置、设备、介质和产品
EP4071683A1 (en) Intelligent computer functionality and visualization for inventory item placement
US10073883B1 (en) Returning query results
CN114549125A (zh) 物品推荐方法及装置、电子设备和计算机可读存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17906870

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 17906870

Country of ref document: EP

Kind code of ref document: A1