WO2019174549A1 - 信息推荐方法和装置 - Google Patents

信息推荐方法和装置 Download PDF

Info

Publication number
WO2019174549A1
WO2019174549A1 PCT/CN2019/077685 CN2019077685W WO2019174549A1 WO 2019174549 A1 WO2019174549 A1 WO 2019174549A1 CN 2019077685 W CN2019077685 W CN 2019077685W WO 2019174549 A1 WO2019174549 A1 WO 2019174549A1
Authority
WO
WIPO (PCT)
Prior art keywords
item
scene
matching
user
items
Prior art date
Application number
PCT/CN2019/077685
Other languages
English (en)
French (fr)
Inventor
陈思聪
于海
Original Assignee
北京京东尚科信息技术有限公司
北京京东世纪贸易有限公司
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 北京京东尚科信息技术有限公司, 北京京东世纪贸易有限公司 filed Critical 北京京东尚科信息技术有限公司
Priority to US16/979,285 priority Critical patent/US11748799B2/en
Publication of WO2019174549A1 publication Critical patent/WO2019174549A1/zh

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
    • G06Q30/00Commerce
    • G06Q30/06Buying, selling or leasing transactions
    • G06Q30/0601Electronic shopping [e-shopping]
    • G06Q30/0623Item investigation
    • G06Q30/0625Directed, with specific intent or strategy
    • G06Q30/0629Directed, with specific intent or strategy for generating comparisons
    • 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
    • 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
    • G06Q30/0201Market modelling; Market analysis; Collecting market data
    • 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
    • G06Q30/06Buying, selling or leasing transactions
    • 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
    • G06Q30/06Buying, selling or leasing transactions
    • G06Q30/0601Electronic shopping [e-shopping]
    • G06Q30/0631Item recommendations
    • 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
    • G06Q30/06Buying, selling or leasing transactions
    • G06Q30/0601Electronic shopping [e-shopping]
    • G06Q30/0641Shopping interfaces
    • G06Q30/0643Graphical representation of items or shoppers

Definitions

  • the present disclosure relates to the field of information processing, and in particular, to an information recommendation method and apparatus.
  • an information recommendation method including: identifying matching item pairs according to historical data to determine an item cross-category relationship; and constructing an item matching scene by using the item cross-category relationship Sorting the matching scenes by using the item scene image and the user scene image; sorting the items in the matching scene according to the feedback behavior of the items in the matching scene; recommending the sorted matching scene to user.
  • the identifying the pair of items according to the historical data includes: the number N ab of being jointly purchased among all users according to the first item a and the second item b, and the first items a and second Determining the matching score of the first item a and the second item b by the expected number of times e bb of the item b being purchased simultaneously; and matching the first item a and the second item b as the matching score exceeds a predetermined threshold Pair of items.
  • the collocation score of the first item a and the second item b is a ratio of a difference between N ab and E ab and f(E ab ), wherein the function f is a predetermined transformation function.
  • the constructing the item matching scenario by using the item across the category relationship comprises: aggregating the item cross-category relationship to aggregate the items having the pair relationship in the same set, thereby Build an item match scene.
  • the sorting the collocation scene by using the item scene portrait and the user scene portrait comprises: generating a combination feature of the user and the scene by using the item scene portrait and the user scene portrait; and performing coefficient pairs according to the combined feature
  • the matching scenes are sorted to preferentially recommend the matching scene with high user interest to the user.
  • the sorting the items in the collocation scene according to the feedback behavior of the user in the collocation scene comprises: determining, according to a feedback behavior of each item in the collocation scene by the user
  • the score of the normalized breakage cumulative gain of the collocation scene is optimized according to the score to improve the ordering of items with high scores and to reduce the ordering of items with low scores.
  • an information recommendation apparatus comprising: a cross-category relationship determining module configured to identify a pair of matching items based on historical data to determine an item cross-category relationship; a building module configured to construct an item matching scene by using the item cross-category relationship; the scene sorting module is configured to sort the matching scene by using an item scene portrait and a user scene portrait; the item sorting module is configured to be configured according to The user sorts the items in the collocation scene by the feedback behavior of the items in the collocation scene; and the recommendation module is configured to recommend the sorted collocation scene to the user.
  • the cross-category relationship determining module is configured to simultaneously purchase the first item a and the second item b in accordance with the number of times Nab of the first item a and the second item b being jointly purchased among all users
  • the expected number of times E ab the matching score of the first item a and the second item b is determined; in the case where the matching score exceeds a predetermined threshold, the first item a and the second item b are used as matching item pairs.
  • the cross-category relationship determining module is further configured to use a ratio of a difference between N ab and E ab and f(E ab ) as a matching score of the first item a and the second item b, wherein
  • the function f is a predetermined transform function.
  • the collocation scene construction module is configured to aggregate the item cross-category relationships to aggregate items having a pair relationship in the same collection to construct an item collocation scene.
  • the scene sorting module is configured to generate a combined feature of the user and the scene using the item scene portrait and the user scene portrait, and sort the matching scene according to the coefficient of the combined feature, so as to have high user interest
  • the matching scenario is recommended to the user first.
  • the item ranking module is configured to determine a corresponding score according to a feedback behavior of the user in each of the collocation scenes, and optimize a normalized breakage cumulative gain of the collocation scene according to the score. In order to improve the ordering of items with high scores, and to reduce the ordering of items with low scores.
  • an information recommendation apparatus comprising: a memory configured to store an instruction; a processor coupled to the memory, the processor being configured to perform an implementation based on the memory stored instruction as described above The method involved in the examples.
  • a computer readable storage medium stores computer instructions that, when executed by a processor, implement a method as recited in any of the above embodiments.
  • FIG. 1 is an exemplary flowchart of an information recommendation method according to an embodiment of the present disclosure
  • FIG. 2 is a diagram showing an example of an article cross-category relationship according to an embodiment of the present disclosure
  • FIG. 3 is an exemplary block diagram of an information recommendation apparatus according to an embodiment of the present disclosure.
  • FIG. 4 is an exemplary block diagram of an information recommendation apparatus according to another embodiment of the present disclosure.
  • the matching scene is constructed by manual means, and the items included in the same matching scene are selected by the operator according to experience. Due to the limited experience of the operators, the items configured in the same collocation scene basically belong to the same category or similar category, and cannot meet the needs of the users. Accordingly, the present disclosure proposes an information recommendation scheme, which can effectively aggregate cross-category collocation products in the same collocation scenario, thereby effectively improving the user experience.
  • FIG. 1 is an exemplary flowchart of an information recommendation method according to an embodiment of the present disclosure.
  • the method steps of the present embodiment can be performed by an information recommendation device.
  • step 101 matching pairs of items are identified based on historical data to determine an item cross-category relationship.
  • recognizing the pair of matching items based on the historical data includes: the number N ab of being jointly purchased among all users according to the first item a and the second item b, and the first item a and the second item b The expected number of times E ab to be purchased simultaneously determines the collocation score of the first item a and the second item b. In the case where the collocation score exceeds a predetermined threshold, the first item a and the second item b are used as matching item pairs.
  • N ab can be obtained directly by the following formula. which is:
  • N ab support(buyers a ⁇ buyers b )
  • buyers a indicates the number of times all users purchase item a
  • buyers b indicates the number of times all users purchase item b
  • N ab represents the number of times the item a and the item b are purchased simultaneously among all users.
  • E ab represents the expected number of times that all users are simultaneously purchased when there is no special relationship between items a and b.
  • the correlation score Score (a, b) is calculated for each candidate pair.
  • the item pair whose score exceeds the threshold is selected as the finally identified matching item pair to ensure the credibility and support of the result.
  • the collocation score of the first item a and the second item b is the ratio of the difference between N ab and E ab to f(E ab ).
  • the function f is a predetermined transform function. which is:
  • Score (a,b) (N ab -E ab )/f(E ab )
  • f(E ab ) is a function of E ab and can be set according to the mining target.
  • the function f can be a root number function, or an nth power function, and the like.
  • the purpose of using the function f is that if the denominator uses E ab directly, it may result in the denominator being too large (the penalty is too strong).
  • the denominator can be placed in a reasonable range, and the penalty effect of the denominator on the entire score is weakened.
  • step 102 the item collocation scene is constructed using the item cross-category relationship.
  • the item collocation scene is constructed by aggregating the item across the category relationships to aggregate the items in the paired relationship in the same collection.
  • the item relationship can be described in a graphical view.
  • Each item can be thought of as a vertex of the graph, and the item cross-category relationship can be considered as the edge of the graph.
  • the corresponding score is the weight of the edge.
  • the collocation scene construction can be performed using an unsupervised learning approach. For example, each item may be initially given a random scene number, and the scene number is continuously propagated and iterated to obtain a scene to which the item with higher reliability belongs.
  • step 103 the matching scenes are sorted using the item scene portrait and the user scene portrait.
  • the combined features of the user and the scene are generated using the item scene portrait and the user scene portrait.
  • the matching scenes are sorted according to the coefficients of the combined features, so that the matching scenarios with high user interest are preferentially recommended to the user.
  • the item scene portrait may include behavior data of the scene to which the item belongs, feedback data of the scene to which the item belongs, quality related data of the item, correlation of the item with the scene, and the like.
  • the user scene portrait may include a user's interest feature for the scene, a user portrait feature, and the like.
  • the item scene portrait and the user scene portrait are first learned by a strong learner to obtain a user-scene combination feature.
  • the weak classifier can be further utilized to score based on the resulting combined features to obtain coefficients for each combined feature. Then, the matching scenes are sorted according to the obtained coefficients.
  • the item scene image and the user scene image are learned by the strong learner, and the weak classifier
  • step 104 the items in the matching scene are sorted according to the feedback behavior of the user on the items in the matching scene, so as to optimize the matching scene.
  • the corresponding score may be determined based on the user's feedback behavior for each item in the collocation scene. According to the score, the NDCG (Normalized Discounted Cumulative Gain) is optimized to match the ranking of the items corresponding to the high score behavior, and the order of the items corresponding to the low score behavior is lowered.
  • NDCG Normalized Discounted Cumulative Gain
  • the corresponding score can be determined according to the user's behavior of clicking, purchasing, searching, etc. on the item. If the user has already purchased the item and the user does not normally purchase it again in the short term, the corresponding score can be lowered to bring the item back. If the user has not purchased the item, but the click, search or pageview of the item is large, indicating that the user has a high degree of interest in the item, the corresponding score can be increased to rank the item forward.
  • the user usually reads from front to back. Therefore, if the items that the user is most interested in can be placed in the front row, the user's purchasing efficiency can be effectively improved, and the browsing time can be reduced. To this end, the items in a scene can be sorted as a whole to optimize. In some embodiments, by ranking the NDCG as an optimization target, the ranking of the items corresponding to the high score behavior is increased, and the ranking of the items corresponding to the low score is reduced.
  • the item list in each scene is personalized and sorted, and the optimization result is obtained.
  • the optimized ranking of items a, b, c is b, a, c.
  • the optimal ranking of items d, e is e, d.
  • the optimal ranking of items h, g, and f is f, g, and h.
  • Table 3 The corresponding scene and item sorting results are shown in Table 3.
  • Sort Scene content 1 A2 (e, d) 2 A1 (b, a, c) 3 A3 (h, g, f)
  • the scenes of interest to the user can be ranked in front, and the cross-category items that the user prefers to purchase in each scene are ranked first. This can effectively enhance the user experience.
  • step 105 the sorted matching scene is recommended to the user.
  • the user experience and the purchase desire are effectively improved by aggregating related or collocated products across the categories and presenting them to the user in an optimized order.
  • scene A including merchandise trekking poles, tents, jackets, goggles
  • scene B including merchandise hot pot bottoms, beef balls, fat cows, hot pot dips
  • each scene is sorted according to the historical behavior of each user.
  • FIG. 3 is an exemplary block diagram of an information recommendation apparatus according to an embodiment of the present disclosure.
  • the information recommendation apparatus includes a cross-category relationship determination module 31, a collocation scene construction module 32, a scene sequencing module 33, an item ranking module 34, and a recommendation module 35.
  • the cross-category relationship determination module 31 is configured to identify matching item pairs based on historical data to determine an item cross-category relationship.
  • the cross-category relationship determination module 31 takes the ratio of the difference between N ab and E ab to f(E ab ) as the collocation score of the first item a and the second item b, wherein the function f is a predetermined transformation function .
  • the function f can be a root number function, or an nth power function, and the like.
  • the collocation scene construction module 32 is configured to construct an item collocation scene using the item cross-category relationship.
  • the collocation scene construction module 32 aggregates the item cross-category relationships to aggregate items having a pair relationship in the same collection to construct an item collocation scene.
  • the collocation scene construction can be performed using an unsupervised learning approach. For example, each item may be initially given a random scene number, and the scene number is continuously propagated and iterated to obtain a scene to which the item with higher reliability belongs.
  • the scene sorting module 33 is configured to sort the matching scenes using the item scene portrait and the user scene portrait.
  • the scene sorting module 33 uses the item scene image and the user scene image to generate a combined feature of the user and the scene, and sorts the matching scene according to the coefficient of the combined feature, so as to preferentially recommend the matching scene with high user interest to the matching scene. user.
  • the item scene portrait and the user scene portrait may be learned by a strong learner to obtain a user-scene combination feature.
  • the weak classifier can be further utilized to score based on the resulting combined features to obtain coefficients for each combined feature. Then, the matching scenes are sorted according to the obtained coefficients.
  • the item ranking module 34 is configured to sort the items in the matching scene according to the feedback behavior of the user on the items in the matching scene to optimize the matching scene.
  • the item ranking module 34 determines a corresponding score according to the feedback behavior of the user for each item in the matching scene, and optimizes the normalized breakage cumulative gain of the matching scene according to the score, so as to improve the ordering of the items with high scores. Reduce the ordering of items with low scores.
  • the list of items in each scene is individually sorted to obtain an optimized result.
  • the recommendation module 35 is configured to recommend the sorted collocation scene to the user.
  • the user experience and the purchase desire are effectively improved by aggregating the related or collocated products across the categories and presenting them to the user in an optimized order.
  • FIG. 4 is an exemplary block diagram of an information recommendation apparatus according to another embodiment of the present disclosure.
  • the information recommendation device includes a memory 41 and a processor 42.
  • Memory 41 is used to store instructions
  • processor 42 is coupled to memory 41
  • processor 42 is configured to perform the methods involved in any of the embodiments of FIG. 1 based on instructions stored in the memory.
  • the information recommendation device further includes a communication interface 43 for performing information interaction with other devices.
  • the apparatus further includes a bus 44, and the processor 42, the communication interface 43, and the memory 41 complete communication with each other via the bus 44.
  • the memory 41 may include a high speed RAM memory, and may also include a non-volatile memory such as at least one disk memory.
  • the memory 41 can also be a memory array.
  • the memory 41 may also be partitioned, and the blocks may be combined into a virtual volume according to certain rules.
  • processor 42 can be a central processing unit CPU, or can be an application specific integrated circuit ASIC, or one or more integrated circuits configured to implement embodiments of the present disclosure.
  • the present disclosure also relates to a computer readable storage medium, wherein the computer readable storage medium stores computer instructions that, when executed by a processor, implement the method of any of the embodiments of FIG.
  • the functional unit modules described above may be implemented as a general purpose processor, a programmable logic controller (PLC), and a digital signal processor (Digital Signal Processor) for performing the functions described in the present disclosure. , referred to as: DSP), Application Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, Discrete hardware components or any suitable combination thereof.
  • DSP digital signal processor
  • ASIC Application Specific Integrated Circuit
  • FPGA Field-Programmable Gate Array
  • the solution provided by the present disclosure can automatically identify cross-category related products in a billion-level mass product, automatically construct a shopping matching scene, and perform scene sorting and recommendation. Covering a large number of products, the speed of updating is far from being achieved by manual operations.
  • the scenario constructed by the present disclosure can effectively cover cross-category goods and can eliminate alternative similar commodities, which is difficult to achieve by traditional methods such as clustering.
  • the present disclosure proposes a novel set of feature combinations and sorting ideas for the shopping scene. Compared with the traditional method, it is more suitable for the scene recommendation, and can achieve better accuracy and effect.
  • a person skilled in the art may understand that all or part of the steps of implementing the above embodiments may be completed by hardware, or may be instructed by a program to execute related hardware, and the program may be stored in a computer readable storage medium.
  • the storage medium mentioned may be a read only memory, a magnetic disk or an optical disk or the like.

Landscapes

  • Business, Economics & Management (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Strategic Management (AREA)
  • Engineering & Computer Science (AREA)
  • Development Economics (AREA)
  • General Physics & Mathematics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Data Mining & Analysis (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种信息推荐方法和装置。该信息推荐装置根据历史数据识别出相搭配的物品对,以确定物品跨类目关系(101),利用物品跨类目关系构建物品搭配场景(102),利用物品场景画像和用户场景画像对搭配场景进行排序(103),根据用户对搭配场景中各物品的反馈行为对搭配场景中的物品进行排序(104),将排序后的搭配场景推荐给用户(105)。该装置通过将跨类目的相关或搭配商品进行聚合,并以优化顺序呈现给用户,从而有效提升了用户体验。

Description

信息推荐方法和装置
相关申请的交叉引用
本申请是以CN申请号为201810199781.X,申请日为2018年3月12日的申请为基础,并主张其优先权,该CN申请的公开内容在此作为整体引入本申请中。
技术领域
本公开涉及信息处理领域,特别涉及一种信息推荐方法和装置。
背景技术
用户在通过电商平台购物时,会存在同时购买归属于不同类目的物品的需求。在相关技术中,通过人工方式将归属于不同类目的物品进行搭配,以构建搭配场景。用户通过搭配场景购买跨类目的物品。
发明内容
根据本公开实施例的第一方面,提供一种信息推荐方法,包括:根据历史数据识别出相搭配的物品对,以确定物品跨类目关系;利用所述物品跨类目关系构建物品搭配场景;利用物品场景画像和用户场景画像对所述搭配场景进行排序;根据用户对所述搭配场景中各物品的反馈行为,对所述搭配场景中的物品进行排序;将排序后的搭配场景推荐给用户。
在一些实施例中,所述根据历史数据识别出相搭配的物品对包括:根据第一物品a和第二物品b在全部用户中被共同购买的次数N ab、以及第一物品a和第二物品b被同时购买的期望次数E ab,确定第一物品a和第二物品b的搭配得分;在所述搭配得分超过预定阈值的情况下,将第一物品a和第二物品b作为相搭配的物品对。
在一些实施例中,所述第一物品a和第二物品b的搭配得分为N ab与E ab的差值与f(E ab)的比值,其中函数f为预定变换函数。
在一些实施例中,所述利用所述物品跨类目关系构建物品搭配场景包括:对所述物品跨类目关系进行聚合处理,以便将具有成对关系的物品聚合在同一个集合中,从而构建出物品搭配场景。
在一些实施例中,所述利用物品场景画像和用户场景画像对所述搭配场景进行排 序包括:利用物品场景画像和用户场景画像,生成用户与场景的组合特征;根据所述组合特征的系数对所述搭配场景进行排序,以便将用户兴趣度高的搭配场景优先推荐给用户。
在一些实施例中,所述根据用户对所述搭配场景中各物品的反馈行为,对所述搭配场景中的物品进行排序包括:根据用户对所述搭配场景中各物品的反馈行为,确定相应的得分;根据所述得分对所述搭配场景的归一化折损累积增益进行优化,以便提升具有高分的物品的排序,降低具有低分的物品的排序。
根据本公开实施例的第二方面,提供一种信息推荐装置,包括:跨类目关系确定模块,被配置为根据历史数据识别出相搭配的物品对,以确定物品跨类目关系;搭配场景构建模块,被配置为利用所述物品跨类目关系构建物品搭配场景;场景排序模块,被配置为利用物品场景画像和用户场景画像对所述搭配场景进行排序;物品排序模块,被配置为根据用户对所述搭配场景中各物品的反馈行为,对所述搭配场景中的物品进行排序;推荐模块,被配置为将排序后的搭配场景推荐给用户。
在一些实施例中,跨类目关系确定模块被配置为根据第一物品a和第二物品b在全部用户中被共同购买的次数N ab、以及第一物品a和第二物品b被同时购买的期望次数E ab,确定第一物品a和第二物品b的搭配得分;在所述搭配得分超过预定阈值的情况下,将第一物品a和第二物品b作为相搭配的物品对。
在一些实施例中,跨类目关系确定模块还被配置为将N ab与E ab的差值与f(E ab)的比值作为所述第一物品a和第二物品b的搭配得分,其中函数f为预定变换函数。
在一些实施例中,搭配场景构建模块被配置为对所述物品跨类目关系进行聚合处理,以便将具有成对关系的物品聚合在同一个集合中,从而构建出物品搭配场景。
在一些实施例中,场景排序模块被配置为利用物品场景画像和用户场景画像,生成用户与场景的组合特征,根据所述组合特征的系数对所述搭配场景进行排序,以便将用户兴趣度高的搭配场景优先推荐给用户。
在一些实施例中,物品排序模块被配置为根据用户对所述搭配场景中各物品的反馈行为,确定相应的得分,根据所述得分对所述搭配场景的归一化折损累积增益进行优化,以便提升具有高分的物品的排序,降低具有低分的物品的排序。
根据本公开实施例的第三方面,提供一种信息推荐装置,包括:存储器,被配置为存储指令;处理器,耦合到存储器,处理器被配置为基于存储器存储的指令执行实现如上述任一实施例涉及的方法。
根据本公开实施例的第四方面,提供一种计算机可读存储介质,其中,计算机可读存储介质存储有计算机指令,指令被处理器执行时实现如上述任一实施例涉及的方法。
通过以下参照附图对本公开的示例性实施例的详细描述,本公开的其它特征及其优点将会变得清楚。
附图说明
为了更清楚地说明本公开实施例或现有技术中的技术方案,下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图仅仅是本公开的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动性的前提下,还可以根据这些附图获得其他的附图。
图1为本公开一个实施例的信息推荐方法的示例性流程图;
图2为本公开一个实施例的物品跨类目关系的示例图;
图3为本公开一个实施例的信息推荐装置的示例性框图;
图4为本公开另一个实施例的信息推荐装置的示例性框图。
具体实施方式
下面将结合本公开实施例中的附图,对本公开实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例仅仅是本公开一部分实施例,而不是全部的实施例。以下对至少一个示例性实施例的描述实际上仅仅是说明性的,决不作为对本公开及其应用或使用的任何限制。基于本公开中的实施例,本领域普通技术人员在没有作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。
除非另外具体说明,否则在这些实施例中阐述的部件和步骤的相对布置、数字表达式和数值不限制本公开的范围。
同时,应当明白,为了便于描述,附图中所示出的各个部分的尺寸并不是按照实际的比例关系绘制的。
对于相关领域普通技术人员已知的技术、方法和设备可能不作详细讨论,但在适当情况下,所述技术、方法和设备应当被视为授权说明书的一部分。
在这里示出和讨论的所有示例中,任何具体值应被解释为仅仅是示例性的,而不是作为限制。因此,示例性实施例的其它示例可以具有不同的值。
应注意到:相似的标号和字母在下面的附图中表示类似项,因此,一旦某一项在一个附图中被定义,则在随后的附图中不需要对其进行进一步讨论。
发明人经过研究发现,在相关技术中,搭配场景是通过人工方式进行构建,同一搭配场景中包括的物品由运营人员根据经验进行挑选。由于运营人员的经验有限,在同一搭配场景中所配置的物品基本上还属于相同类目或近似类目,并不能满足用户的需求。据此,本公开提出一种信息推荐方案,可有效地将跨类目的搭配商品聚合在同一个搭配场景中,有效提升了用户体验。
图1为本公开一个实施例的信息推荐方法的示例性流程图。在一些实施例中,本实施例的方法步骤可由信息推荐装置执行。
在步骤101中,根据历史数据识别出相搭配的物品对,以确定物品跨类目关系。
在一些实施例中,根据历史数据识别出相搭配的物品对包括:根据第一物品a和第二物品b在全部用户中被共同购买的次数N ab、以及第一物品a和第二物品b被同时购买的期望次数E ab,确定第一物品a和第二物品b的搭配得分。在搭配得分超过预定阈值的情况下,将第一物品a和第二物品b作为相搭配的物品对。
N ab可直接通过下列公式得到。即:
N ab=support(buyers a∩buyers b)
在上述公式中,buyers a表示全部用户购买物品a的次数,buyers b表示全部用户购买物品b的次数。N ab表示全部用户中同时购买物品a和物品b的次数。
假设用户的每次购买行为是相互独立的,且服从伯努利分布。当物品a、b不具备相关关系时,可通过伯努利分布的概率计算E ab。即E ab代表了当物品a、b无特殊关系时,被所有用户同时购买的期望次数。
例如,在一定时间段内,所有被用户共同购买的物品对作为相搭配的候选对。对每个候选对计算相关得分Score (a,b)。选取得分超过阈值的物品对作为最终识别出的相搭配的物品对,以保证结果的可信度及支持度。
在一些实施例中,第一物品a和第二物品b的搭配得分为N ab与E ab之差与f(E ab)的比值。函数f为预定变换函数。即:
Score (a,b)=(N ab-E ab)/f(E ab)
在上述公式中,f(E ab)为E ab的函数,可根据挖掘目标自行设定。例如,函数f可以为开根号函数,或者开n次方函数等。使用函数f的目的是,若分母直接使用E ab,可能导致分母过大(惩罚过强)。通过对E ab进行诸如开根号的变换处理,可以使分 母处于一个合理的范围内,减弱分母对整个得分的惩罚效果。
在步骤102中,利用物品跨类目关系构建物品搭配场景。
在一些实施例中,通过对物品跨类目关系进行聚合处理,以便将具有成对关系的物品聚合在同一个集合中,从而构建出物品搭配场景。
在一些实施例中,在得到物品之间的跨类目关系后,可以以图的观点描述物品关系。每个物品可以视为图的一个顶点,物品跨类目关系可以视为图的边。相应的得分为边的权重。通过在全体物品集合中挖掘物品集合组成的子图,以便将物品之间的相关关系聚合成社区形式,即物品搭配场景。
如图2所示,对于物品a、b、c、d和e,若两个物品之间存在相关关系,则用边(短线)表示,这两个物品之间的得分为相应边的权重。当然,若两个物品之间不相关,如物品a和e,则其两者之间就不会有边。
在一些实施例中,可利用无监督学习的方法进行搭配场景构建。例如,可初始赋予每个物品一个随机的场景编号,通过将这个场景编号进行不断传播迭代,来取得可信度较高的物品所属场景。
需要说明的是,由于无监督学习自身并不是本公开的发明点所在,因此这里不展开描述。
在得到场景之后,通过人工命名的方式,即可得到线上可用的购物搭配场景。在本公开中,人工干预仅为场景上线前的人工命名及审核。场景的挖掘、过滤、物品选择均无需人工干预,此外还可以定期进行自动更新。
例如,对于8个物品a、b、c、d、e、f、g和h,通过上述步骤101可知,相搭配的物品对为a→b、a→c、d→e、f→h、g→h。通过聚合处理,将这8个物品分成3个群组。即,群组A1(a、b、c)、群组A2(d、e)和群组A3(f、g、h)。从而将群组1、群组2和群组3作为三个不同的场景。
例如,对于“户外出行”这一场景,通过上述处理,可包括跨类目的物品集合。如表1所示。
场景所涵盖物品 物品所属类目
背包 箱包类目
登山杖 户外装备类目
冲锋衣 服饰类目
登山鞋 鞋靴类目
旅行食品 食品类目
…… ……
表1
又例如,对于“火锅大餐”这一场景,通过上述处理,可包括跨类目的物品集合。如表2所示。
场景所涵盖物品 物品所属类目
涮锅 厨具类目
食材 生鲜类目
蘸料 调味品类目
佐餐酒 酒类目
…… ……
表2
在步骤103中,利用物品场景画像和用户场景画像对搭配场景进行排序。
在一些实施例中,利用物品场景画像和用户场景画像,生成用户与场景的组合特征。根据组合特征的系数对搭配场景进行排序,以便将用户兴趣度高的搭配场景优先推荐给用户。
例如,物品场景画像可包括物品所属场景的行为数据、物品所属场景的反馈数据、物品的质量相关数据、商品与场景的相关性等信息。用户场景画像可包括用户对场景的兴趣特征、用户画像特征等。
在一些实施例中,先通过强学习器对物品场景画像和用户场景画像进行学习,以得到用户-场景的组合特征。此外,可进一步利用弱分类器根据所得到的组合特征进行打分,以便得到各组合特征的系数。进而根据所得到的系数对搭配场景进行排序。通过将强学习器与弱分类器相结合,既能拥有强学习器的准确度,又能拥有弱分类器的可解释性,各个组合特征的内容及关系直观可视。本公开相比于相关技术,能够从数据中自动学习用户与物品之间的组合特征,更适应购物场景频繁更新的特点。
例如,对于三个场景A1(a、b、c)、A2(d、e)和A3(f、g、h),通过强学习器对物品场景画像和用户场景画像进行学习,以及弱分类器的处理,发现在这三个场景中,用户对场景A2最感兴趣,对场景A3的兴趣度最低。因此可根据学习和分类 处理结果,将场景A1、A2和A3排序为A2、A1和A3。
需要说明的是,由于强学习器和弱分类器自身并不是本公开的发明点所在,因此这里不展开说明。
在步骤104中,根据用户对搭配场景中各物品的反馈行为,对搭配场景中的物品进行排序,以便对搭配场景进行优化。
在一些实施例中,可根据用户对搭配场景中各物品的反馈行为,确定相应的得分。根据得分优化搭配场景的NDCG(Normalized Discounted Cumulative Gain,归一化折损累积增益),以便将高分行为对应物品的排序提升,将低分行为对应物品的排序降低。
例如,可根据用户对物品的点击、购买、搜索等行为确定相应的得分。若用户已经购买过该物品,而用户通常不会在短期内再次购买,因此可将相应的得分降低,以便将该物品向后排。若用户还未购买该物品,但是对该物品的点击、搜索或浏览量较大,表明用户对该物品的兴趣度较高,因此可将相应的得分提升,以便将该物品向前排。
通常,对于推荐信息,用户总是从前向后阅读。因此,如果能将用户最感兴趣的物品往前排,则可有效提高用户的购买效率,减少浏览时间。为此,可将一个场景内的物品整体排序进行优化。在一些实施例中,可通过将NDCG作为优化目标,将高分行为对应物品的排序提升,将低分行为对应物品的排序降低。
例如,对于上述的三个场景A1(a、b、c)、A2(d、e)和A3(f、g、h),通过对搭配场景进行排序后发现,用户对场景A2的兴趣度最高,对场景A3的兴趣度最低。因此将场景A1、A2和A3排序为A2、A1和A3。
接下来,以NDCG作为优化目标,分别对每个场景中的物品列表进行个性化排序,从而得到优化结果。例如,对于场景A1来说,物品a、b、c的优化排名为b、a、c。对于场景A2来说,物品d、e的优化排名为e、d。对于场景A3来说,物品h、g、f的优化排名为f、g、h。相应的场景及物品排序结果如表3所示。
排序 场景内容
1 A2(e、d)
2 A1(b、a、c)
3 A3(h、g、f)
表3
通过上述处理,可将用户感兴趣的场景排在前面,将各场景中用户更喜欢购买的跨类目物品排在前面。从而可有效提升用户体验。
需要说明的是,由于NDCG优化自身并不是本公开的发明点所在,因此这里不展开说明。
在步骤105中,将排序后的搭配场景推荐给用户。
在本公开上述实施例提供的信息推荐方法中,通过将跨类目的相关或搭配商品进行聚合,并以优化顺序呈现给用户,从而有效提升了用户体验和购买欲。
下面通过一个具体示例对本公开进行说明。
首先,识别商品之间的跨类目关系。例如,登山杖与帐篷搭配,冲锋衣与护目镜搭配,火锅底料与牛肉丸搭配,肥牛与火锅蘸料搭配,等等。其次,把得到的跨类目关系进行聚合处理。例如,得到了场景A(包含商品登山杖、帐篷、冲锋衣、护目镜),场景B(包含商品火锅底料、牛肉丸、肥牛、火锅蘸料)以及其它场景。至于户外出行和火锅大餐这两个名字,可由人工来取名。接下来,根据每个用户的历史行为,对各场景进行排序。例如,用户a喜欢登山,不喜欢火锅,就把与登山相关的场景排第一个,将与火锅相关的场景尽可能往后排。最后,根据每个用户的行为,对场景内的商品进行排序,比如用户a更喜欢买冲锋衣,并且已经买过登山杖了,那么登山这个场景里,就需要把冲锋衣排前面,登山杖排后面。由此,可将用户最感兴趣场景的跨类目信息提供给用户,从而提升了用户体验。
图3为本公开一个实施例的信息推荐装置的示例性框图。如图3所示,该信息推荐装置包括跨类目关系确定模块31、搭配场景构建模块32、场景排序模块33、物品排序模块34和推荐模块35。
跨类目关系确定模块31被配置为根据历史数据识别出相搭配的物品对,以确定物品跨类目关系。
在一些实施例中,跨类目关系确定模块31根据第一物品a和第二物品b在全部用户中被共同购买的次数N ab、以及第一物品a和第二物品b被同时购买的期望次数E ab,确定第一物品a和第二物品b的搭配得分。在搭配得到超过预定阈值的情况下,跨类目关系确定模块31将第一物品a和第二物品b作为相搭配的物品对。
在一些实施例中,跨类目关系确定模块31将N ab与E ab之差与f(E ab)的比值作为第一物品a和第二物品b的搭配得分,其中函数f为预定变换函数。例如,函数f可以为开根号函数,或者开n次方函数等。
搭配场景构建模块32被配置为利用物品跨类目关系构建物品搭配场景。
在一些实施例中,搭配场景构建模块32对物品跨类目关系进行聚合处理,以便将具有成对关系的物品聚合在同一个集合中,从而构建出物品搭配场景。
在一些实施例中,可利用无监督学习的方法进行搭配场景构建。例如,可初始赋予每个物品一个随机的场景编号,通过将这个场景编号进行不断传播迭代,来取得可信度较高的物品所属场景。
场景排序模块33被配置为利用物品场景画像和用户场景画像对搭配场景进行排序。
在一些实施例中,场景排序模块33利用物品场景画像和用户场景画像,生成用户与场景的组合特征,根据组合特征的系数对搭配场景进行排序,以便将用户兴趣度高的搭配场景优先推荐给用户。
在一些实施例中,可先通过强学习器对物品场景画像和用户场景画像进行学习,以得到用户-场景的组合特征。此外,可进一步利用弱分类器根据所得到的组合特征进行打分,以便得到各组合特征的系数。进而根据所得到的系数对搭配场景进行排序。通过将强学习器与弱分类器相结合,既能拥有强学习器的准确度,又能拥有弱分类器的可解释性,各个组合特征的内容及关系直观可视。本公开相比于相关技术,能够从数据中自动学习用户与物品之间的组合特征,更适应购物场景频繁更新的特点。
物品排序模块34被配置为根据用户对搭配场景中各物品的反馈行为,对搭配场景中的物品进行排序,以便对搭配场景进行优化。
在一些实施例中,物品排序模块34根据用户对搭配场景中各物品的反馈行为确定相应的得分,根据得分优化搭配场景的归一化折损累积增益,以便提升具有高分的物品的排序,降低具有低分的物品的排序。
在一些实施例中,以NDCG作为优化目标,分别对每个场景中的物品列表进行个性化排序,从而得到优化结果。
推荐模块35被配置为将排序后的搭配场景推荐给用户。
在本公开信息推荐装置的上述实施例中,通过将跨类目的相关或搭配商品进行聚合,并以优化顺序呈现给用户,从而有效提升了用户体验和购买欲。
图4为本公开另一个实施例的信息推荐装置的示例性框图。如图4所示,信息推荐装置包括存储器41和处理器42。
存储器41用于存储指令,处理器42耦合到存储器41,处理器42被配置为基于 存储器存储的指令执行实现如图1中任一实施例涉及的方法。
如图4所示,该信息推荐装置还包括通信接口43,用于与其它设备进行信息交互。同时,该装置还包括总线44,处理器42、通信接口43、以及存储器41通过总线44完成相互间的通信。
存储器41可以包含高速RAM存储器,也可还包括非易失性存储器(non-volatile memory),例如至少一个磁盘存储器。存储器41也可以是存储器阵列。存储器41还可能被分块,并且块可按一定的规则组合成虚拟卷。
此外,处理器42可以是一个中央处理器CPU,或者可以是专用集成电路ASIC,或者是被配置成实施本公开实施例的一个或多个集成电路。
本公开同时还涉及一种计算机可读存储介质,其中计算机可读存储介质存储有计算机指令,指令被处理器执行时实现如图1中任一实施例涉及的方法。
在一些实施例中,上述的功能单元模块可以实现为用于执行本公开所描述功能的通用处理器、可编程逻辑控制器(Programmable Logic Controller,简称:PLC)、数字信号处理器(Digital Signal Processor,简称:DSP)、专用集成电路(Application Specific Integrated Circuit,简称:ASIC)、现场可编程门阵列(Field-Programmable Gate Array,简称:FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件或者其任意适当组合。
通过实施本公开所提供的方案,可得到以下有益效果中的至少一项:
(1)本公开所提供的方案能够在亿级别海量商品中自动识别跨品类相关商品,自动构建成购物搭配场景,并进行场景排序和推荐。覆盖商品之多,更新速度之快,是人工运营远远无法达到的。
(2)本公开方案所构建的商品相关关系能够涵盖各个维度的商品相关信息。相比之下,人工运营受限于运营人员的知识及经验,难以达到本公开方案的精度及多样性。
(3)在本公开方案构建商品搭配场景的过程中,所需的人工干预仅为新场景的命名及上线审核,极大地节约了人力成本。
(4)本公开所构建的场景能够有效涵盖跨类目商品,能够剔除替代性的相似商品,这是聚类等传统方法难以实现的。
(5)本公开针对购物场景,提出了一整套新颖的特征组合及排序思路。相对于传统方法,更适应场景化推荐,能够取得更好的准确度和效果。
本领域普通技术人员可以理解实现上述实施例的全部或部分步骤可以通过硬件来完成,也可以通过程序来指令相关的硬件完成,所述的程序可以存储于一种计算机可读存储介质中,上述提到的存储介质可以是只读存储器,磁盘或光盘等。
本公开的描述是为了示例和描述起见而给出的,而并不是无遗漏的或者将本公开限于所公开的形式。很多修改和变化对于本领域的普通技术人员而言是显然的。选择和描述实施例是为了更好说明本公开的原理和实际应用,并且使本领域的普通技术人员能够理解本公开从而设计适于特定用途的带有各种修改的各种实施例。

Claims (14)

  1. 一种信息推荐方法,包括:
    根据历史数据识别出相搭配的物品对,以确定物品跨类目关系;
    利用所述物品跨类目关系构建物品搭配场景;
    利用物品场景画像和用户场景画像对所述搭配场景进行排序;
    根据用户对所述搭配场景中各物品的反馈行为,对所述搭配场景中的物品进行排序;
    将排序后的搭配场景推荐给用户。
  2. 根据权利要求1所述的信息推荐方法,其中,所述根据历史数据识别出相搭配的物品对包括:
    根据第一物品a和第二物品b在全部用户中被共同购买的次数N ab、以及第一物品a和第二物品b被同时购买的期望次数E ab,确定第一物品a和第二物品b的搭配得分;
    在所述搭配得分超过预定阈值的情况下,将第一物品a和第二物品b作为相搭配的物品对。
  3. 根据权利要求2所述的信息推荐方法,其中,
    所述第一物品a和第二物品b的搭配得分为N ab与E ab的差值与f(E ab)的比值,其中函数f为预定变换函数。
  4. 根据权利要求1所述的信息推荐方法,其中,所述利用所述物品跨类目关系构建物品搭配场景包括:
    对所述物品跨类目关系进行聚合处理,以便将具有成对关系的物品聚合在同一个集合中,从而构建出物品搭配场景。
  5. 根据权利要求1所述的信息推荐方法,其中,所述利用物品场景画像和用户场景画像对所述搭配场景进行排序包括:
    利用物品场景画像和用户场景画像,生成用户与场景的组合特征;
    根据所述组合特征的系数对所述搭配场景进行排序,以便将用户兴趣度高的搭配场景优先推荐给用户。
  6. 根据权利要求1-5中任一项所述的信息推荐方法,其中,所述根据用户对所述搭配场景中各物品的反馈行为,对所述搭配场景中的物品进行排序包括:
    根据用户对所述搭配场景中各物品的反馈行为,确定相应的得分;
    根据所述得分对所述搭配场景的归一化折损累积增益进行优化,以便提升具有高分的物品的排序,降低具有低分的物品的排序。
  7. 一种信息推荐装置,包括:
    跨类目关系确定模块,被配置为根据历史数据识别出相搭配的物品对,以确定物品跨类目关系;
    搭配场景构建模块,被配置为利用所述物品跨类目关系构建物品搭配场景;
    场景排序模块,被配置为利用物品场景画像和用户场景画像对所述搭配场景进行排序;
    物品排序模块,被配置为根据用户对所述搭配场景中各物品的反馈行为,对所述搭配场景中的物品进行排序;
    推荐模块,被配置为将排序后的搭配场景推荐给用户。
  8. 根据权利要求7所述的信息推荐装置,其中,
    跨类目关系确定模块被配置为根据第一物品a和第二物品b在全部用户中被共同购买的次数N ab、以及第一物品a和第二物品b被同时购买的期望次数E ab,确定第一物品a和第二物品b的搭配得分;在所述搭配得分超过预定阈值的情况下,将第一物品a和第二物品b作为相搭配的物品对。
  9. 根据权利要求8所述的信息推荐装置,其中,
    跨类目关系确定模块还被配置为将N ab与E ab的差值与f(E ab)的比值作为所述第一物品a和第二物品b的搭配得分,其中函数f为预定变换函数。
  10. 根据权利要求7所述的信息推荐装置,其中,
    搭配场景构建模块被配置为对所述物品跨类目关系进行聚合处理,以便将具有成对关系的物品聚合在同一个集合中,从而构建出物品搭配场景。
  11. 根据权利要求7所述的信息推荐装置,其中,
    场景排序模块被配置为利用物品场景画像和用户场景画像,生成用户与场景的组合特征,根据所述组合特征的系数对所述搭配场景进行排序,以便将用户兴趣度高的搭配场景优先推荐给用户。
  12. 根据权利要求7-11中任一项所述的信息推荐装置,其中,
    物品排序模块被配置为根据用户对所述搭配场景中各物品的反馈行为,确定相应的得分,根据所述得分对所述搭配场景的归一化折损累积增益进行优化,以便提升具有高分的物品的排序,降低具有低分的物品的排序。
  13. 一种信息推荐装置,包括:
    存储器,被配置为存储指令;
    处理器,耦合到存储器,处理器被配置为基于存储器存储的指令执行实现如权利要求1-6中任一项的方法。
  14. 一种计算机可读存储介质,其中,计算机可读存储介质存储有计算机指令,指令被处理器执行时实现如权利要求1-6中任一项的方法。
PCT/CN2019/077685 2018-03-12 2019-03-11 信息推荐方法和装置 WO2019174549A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/979,285 US11748799B2 (en) 2018-03-12 2019-03-11 Method, medium, and system for information recommendation

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810199781.XA CN110264291B (zh) 2018-03-12 2018-03-12 信息推荐方法和装置
CN201810199781.X 2018-03-12

Publications (1)

Publication Number Publication Date
WO2019174549A1 true WO2019174549A1 (zh) 2019-09-19

Family

ID=67907317

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/077685 WO2019174549A1 (zh) 2018-03-12 2019-03-11 信息推荐方法和装置

Country Status (3)

Country Link
US (1) US11748799B2 (zh)
CN (1) CN110264291B (zh)
WO (1) WO2019174549A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113194155A (zh) * 2021-05-20 2021-07-30 青岛海信智慧生活科技股份有限公司 一种物联网中场景智能推荐的方法及装置

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110807691B (zh) * 2019-10-31 2022-03-04 深圳市云积分科技有限公司 一种跨商品品类的商品推荐方法和装置
CN113052646A (zh) * 2019-12-27 2021-06-29 阿里巴巴集团控股有限公司 数据处理系统、方法、装置及电子设备
CN113127728A (zh) * 2020-01-16 2021-07-16 北京沃东天骏信息技术有限公司 一种处理物品场景图的方法和装置
CN111798286A (zh) * 2020-04-26 2020-10-20 北京沃东天骏信息技术有限公司 物品搭配方法、物品搭配模型的构建方法和计算机
CN113744011A (zh) * 2020-06-17 2021-12-03 北京沃东天骏信息技术有限公司 物品搭配方法和物品搭配装置
CN113763111A (zh) * 2021-02-10 2021-12-07 北京沃东天骏信息技术有限公司 物品搭配方法、装置及存储介质
CN115880037B (zh) * 2023-03-03 2023-05-12 量子数科科技有限公司 基于多项目规划整合分析的商品推荐方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103208073A (zh) * 2012-01-17 2013-07-17 阿里巴巴集团控股有限公司 获得推荐商品信息及提供商品信息的方法、装置
CN104268155A (zh) * 2014-09-02 2015-01-07 百度在线网络技术(北京)有限公司 用户移动端个性化推荐的及时反馈优化方法和装置
CN105427125A (zh) * 2015-10-29 2016-03-23 电子科技大学 基于商品网络连接图的商品簇划分方法
CN107256513A (zh) * 2017-06-19 2017-10-17 数据堂(北京)科技股份有限公司 一种对象的搭配推荐方法及装置

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7720723B2 (en) * 1998-09-18 2010-05-18 Amazon Technologies, Inc. User interface and methods for recommending items to users
US7743059B2 (en) * 2007-03-30 2010-06-22 Amazon Technologies, Inc. Cluster-based management of collections of items
US7779040B2 (en) * 2007-09-27 2010-08-17 Amazon Technologies, Inc. System for detecting associations between items
JP5967553B2 (ja) * 2014-06-16 2016-08-10 インターナショナル・ビジネス・マシーンズ・コーポレーションInternational Business Machines Corporation 店舗内又は店舗間における顧客の購買行動を推定する方法、並びに、そのコンピュータ・システム及びコンピュータ・プログラム

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103208073A (zh) * 2012-01-17 2013-07-17 阿里巴巴集团控股有限公司 获得推荐商品信息及提供商品信息的方法、装置
CN104268155A (zh) * 2014-09-02 2015-01-07 百度在线网络技术(北京)有限公司 用户移动端个性化推荐的及时反馈优化方法和装置
CN105427125A (zh) * 2015-10-29 2016-03-23 电子科技大学 基于商品网络连接图的商品簇划分方法
CN107256513A (zh) * 2017-06-19 2017-10-17 数据堂(北京)科技股份有限公司 一种对象的搭配推荐方法及装置

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113194155A (zh) * 2021-05-20 2021-07-30 青岛海信智慧生活科技股份有限公司 一种物联网中场景智能推荐的方法及装置

Also Published As

Publication number Publication date
US11748799B2 (en) 2023-09-05
CN110264291B (zh) 2024-06-18
CN110264291A (zh) 2019-09-20
US20200402137A1 (en) 2020-12-24

Similar Documents

Publication Publication Date Title
WO2019174549A1 (zh) 信息推荐方法和装置
US11216861B2 (en) Color based social networking recommendations
CN110458627B (zh) 一种面向用户动态偏好的商品序列个性化推荐方法
CN103246980B (zh) 信息输出方法及服务器
CN107451894B (zh) 数据处理方法、装置和计算机可读存储介质
CN102902691B (zh) 推荐方法及系统
CN104866474B (zh) 个性化数据搜索方法及装置
CN109087177A (zh) 向目标用户推荐商品的方法、装置及计算机可读存储介质
CN110020128B (zh) 一种搜索结果排序方法及装置
US9727620B2 (en) System and method for item and item set matching
CN112015998A (zh) 一种基于用户画像的商品推荐方法
CN108664564A (zh) 一种基于物品内容特征的改进协同过滤推荐方法
CN111681084B (zh) 一种基于社交关系影响因素的电商平台推荐方法
CN109635198A (zh) 在商品展示平台上呈现用户搜索结果的方法、装置、介质及电子设备
KR102109338B1 (ko) 고객 맞춤형 의류 분석 및 추천 시스템
CN108122126B (zh) 关联规则的扩充方法、应用其的装置及电脑可读取介质
US11587028B2 (en) Computer system for optimizing garment inventory of retailer based on shapes of users
CN106682923A (zh) 商品调整方法及系统
CN112132660A (zh) 商品推荐方法、系统、设备及存储介质
CN112424812A (zh) 计算电子处理平台上的用户特定物品相关值
Vartak et al. CHIC: a combination-based recommendation system
CN113763089A (zh) 物品推荐方法及装置、计算机可存储介质
KR20220001616A (ko) 통합 고객 식별자 생성을 기반으로 하는 고객 빅데이터 구축 방법, 장치 및 시스템
KR20220001617A (ko) 고객 빅데이터를 활용한 상품 추천 방법, 장치 및 시스템
CN108182288A (zh) 一种基于人工免疫的推荐方法

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: 19768454

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205 DATED 20/01/2021)

122 Ep: pct application non-entry in european phase

Ref document number: 19768454

Country of ref document: EP

Kind code of ref document: A1