WO2017128973A1 - 用户行为数据的处理方法和装置 - Google Patents
用户行为数据的处理方法和装置 Download PDFInfo
- Publication number
- WO2017128973A1 WO2017128973A1 PCT/CN2017/071241 CN2017071241W WO2017128973A1 WO 2017128973 A1 WO2017128973 A1 WO 2017128973A1 CN 2017071241 W CN2017071241 W CN 2017071241W WO 2017128973 A1 WO2017128973 A1 WO 2017128973A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- behavior data
- time
- factor
- generated
- user
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/2866—Architectures; Arrangements
- H04L67/30—Profiles
- H04L67/306—User profiles
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/50—Network services
- H04L67/535—Tracking the activity of the user
Definitions
- the present application relates to the field of Internet technologies, and in particular, to a method and an apparatus for processing user behavior data.
- Search engines often use machine learning methods to optimize the ranking results of multiple search results by learning historical log data. In order to obtain better search results, historical data from the past two weeks, one month or even longer is often used. Predict user preferences for search results.
- search engines usually treat all historical data equally when optimizing the ranking results of search results through historical log data, for example, a behavior log 10 days ago and a historical data 1 day ago, search The engine does not distinguish the importance of historical data when predicting user preferences according to the behavior log. All historical data is predicted by the machine learning method according to the same importance degree, so that the search engine cannot effectively capture the user's interest change, and cannot Meet the needs of users.
- the purpose of the present application is to solve at least one of the above technical problems to some extent.
- the first object of the present application is to provide a method for processing user behavior data, which can improve the timeliness of user behavior data, and can more effectively obtain user preferences and interests, and provide users with more satisfaction. Search results of user needs have improved the user experience.
- a second object of the present application is to provide a processing apparatus for user behavior data.
- a method for processing user behavior data includes the following steps: acquiring behavior data of a user, and acquiring a time for generating the behavior data; The behavior of the behavior data and the current time to calculate the behavior The time-dependent factor and/or periodic factor corresponding to the data; the behavior data is adjusted according to the time-effect factor and/or the periodic factor.
- the processing method of the embodiment of the present application combines time decay and periodic similarity to assign different importance levels to the user's behavior data, and optimizes the ranking results of multiple search results through machine learning methods by using behavior data of different importance levels, thereby improving the ranking result.
- the timeliness of user behavior data can more effectively obtain the user's preferences and interests, provide users with more search results that meet the user's needs, and improve the user experience.
- a device for processing user behavior data includes: an obtaining module, configured to acquire behavior data of a user, and acquire a time for generating the behavior data; And a method for calculating a time-dependent factor and/or a periodic factor corresponding to the behavior data according to the time and current time at which the behavior data is generated; and a processing module, configured to use the time-based factor and/or the periodic factor The behavior data is adjusted.
- the processing device of the embodiment of the present application combines time decay and periodic similarity to assign different importance levels to the user's behavior data, and optimizes the ranking results of multiple search results through machine learning methods by using behavior data of different importance levels, thereby improving the ranking result.
- the timeliness of user behavior data can more effectively obtain the user's preferences and interests, provide users with more search results that meet the user's needs, and improve the user experience.
- FIG. 1 is a flowchart of a method of processing user behavior data according to an embodiment of the present application
- FIG. 2 is a flow chart of a method for processing user behavior data according to an embodiment of the present invention
- FIG. 3 is a flow chart of a method for processing user behavior data according to another embodiment of the present invention.
- FIG. 4 is a schematic structural diagram of a device for processing user behavior data according to an embodiment of the present application.
- FIG. 1 is a flow chart of a method of processing user behavior data in accordance with one embodiment of the present application.
- the processing method includes:
- S110 Acquire behavior data of the user, and obtain time for generating behavior data.
- the format of the behavior data is generally as shown in the following table, where features is a feature in the behavior data, and label is the user's click behavior (divided into two parts by ⁇ , the first part is 1 for user click, and the second is 1 for The user did not click), date is the time when the click behavior was generated.
- S120 Calculate a timeliness factor and/or a periodicity factor corresponding to the behavior data according to the time at which the behavior data is generated and the current time.
- some user behaviors are cyclical. For example, office worker B often buys ingredients online during the weekend because user B has time to cook on weekends. So although Monday is very close to the last weekend, user B's behavior is still very different. Therefore, when using historical data learning, it is necessary to consider the influence of periodicity, and give historical data different degrees of importance according to how close the historical data is to the current periodicity in a certain dimension.
- the time difference between the time at which the behavior data is generated and the current time is calculated, and the time-effect factor Z is calculated according to a specific function.
- the periodicity factor C is calculated by a specific function according to the time at which the behavior data is generated and the current time.
- the behavior data may be first input to the data segmentation device for segmentation, and the behavior data is divided into a plurality of small behavior data. Piece.
- the data segmentation device determines how many copies of the behavior data are divided according to the size of the behavior data and the amount of data that can be calculated by each computer that processes the behavior data. Since the data segmentation device performs the segmentation of the behavior data itself in a distributed environment, the resources of the plurality of computing devices can be fully utilized to accelerate the segmentation action.
- each small behavior data block is distributed to each computer device, and multiple computer devices can simultaneously process the behavior data block in parallel, and the behavior obtained by any two computer devices is obtained. Data blocks do not overlap.
- S130 Adjust behavior data according to a timeliness factor and/or a periodicity factor.
- the user's behavior data may be adjusted only according to the time-effect factor Z, or only the user's line according to the periodic factor C.
- the data is adjusted, or the user's behavior data can be adjusted according to the time-effect factor Z and the periodic factor C, that is, the user's behavior data is first adjusted according to the time-effect factor Z, and then the user is adjusted according to the periodic factor C.
- the behavioral data is adjusted.
- the method for processing user behavior data in the embodiment of the present application combines time decay and periodic similarity to assign different importance levels to the user's behavior data, and optimizes the ranking of multiple search results through machine learning methods by using behavior data of different importance levels.
- the timeliness of the user behavior data is improved, the user's preferences and interests are more effectively obtained, and the user is provided with search results that more satisfy the user's needs, thereby improving the user experience.
- FIG. 2 is a flowchart of a method for processing user behavior data according to an embodiment of the present invention
- FIG. 3 is a flowchart of a method for processing user behavior data according to another embodiment of the present invention, as shown in FIG. 2 and FIG.
- the processing of behavior data includes:
- the time decay function is an exponential function or a power function.
- S240 Calculate a date difference between the ith day and the current jth day of the generated behavior data.
- the time at which the behavior data is generated and the current time belong to the i-th day and the j-th day in the preset time period, wherein the preset time period has N days, and i and j are positive integers less than or equal to N.
- the periodic factor is generated by the following formula:
- di is the ith day when the behavior data is generated and dj is the current jth day.
- S260 Generate second behavior data according to the periodicity factor, the behavior data, and the first behavior data.
- generating the second behavior data according to the periodicity factor, the behavior data, and the first behavior data specifically includes:
- S262. Generate second behavior data according to the first behavior data and the third behavior data. Wherein the first behavior data and the third behavior data are added or multiplied to generate second behavior data.
- the click behavior label of the user in the user behavior data and the time date of the click behavior are input, and the behavior data of the user is weighted according to the time decay rule. It should be understood that after segmenting the user's behavior data, since only the user's click behavior label and the time date of the click behavior are required to be input, the data processing amount of each computing machine can be greatly reduced. Let t 0 be the current date and t be the date from which the behavior data is generated. Therefore, t 0 -t can measure how far the user's behavior data is from the current date, that is, the distance.
- time decay function f(x) is a decreasing function, ie f(x) is a decreasing function on [0, + ⁇ ), then f(t 0 -t) constitutes a time decay to weight the behavior data.
- the behavioral data weight one day ago is f(1)
- the behavioral data weight three days ago is f(3), and so on. Since f(x) is a decreasing function, f(1)>f(3), the weight of the behavior data that is larger from the current time interval is smaller.
- the current time and the time at which the behavior data is generated may be in days, that is, the current date and the date on which the behavior data is generated, or may be in hours or minutes, etc.
- the description is made only in days, and the weight of the behavior data is calculated in units of hours or minutes and the weight of the behavior data is calculated in units of days. To avoid redundancy, this is not the case. superfluous.
- calculating the difference between the date when the behavior data is generated and the current date is obtained by subtracting the date of the date in the behavior data from the current date to obtain the time difference, and then selecting the exponential function or the power function to calculate the time-effect factor Z, such as aging
- the periodic information is only described in the first few days of the week. Any other periodic unit can be measured in a similar way, for example, 24 hours a day, a change in the solar terms, a rise and fall in temperature, etc. Sexual information.
- label 2 is calculated based on the selected superposition relationship.
- the superposition relationship may be plus or multiplied, and if the superposition relationship is plus, then:
- the sorted result of the plurality of search results is optimized by the machine learning method using the obtained adjusted behavior data label 2 .
- the method for processing user behavior data in the embodiment of the present application firstly processes the behavior data in time, and then periodically processes the behavior data after the time-sensitive processing, thereby passing the machine learning method according to the behavior data after the periodic processing.
- the present application also proposes a processing apparatus for user behavior data.
- FIG. 4 is a schematic structural diagram of a device for processing user behavior data according to an embodiment of the present application.
- the processing device includes an acquisition module 100, a calculation module 200, and a processing module 300.
- the obtaining module 100 is configured to acquire behavior data of the user, and acquire behavior data. time.
- the calculation module 200 is configured to calculate a timeliness factor and/or a periodicity factor corresponding to the behavior data according to the time at which the behavior data is generated and the current time.
- the processing module 300 is configured to adjust the behavior data according to the timeliness factor and/or the periodicity factor.
- the device for processing user behavior data in the embodiment of the present application combines time decay and periodic similarity to assign different importance levels to the user's behavior data, and optimizes the ranking of multiple search results through machine learning methods by using behavior data of different importance levels.
- the timeliness of the user behavior data is improved, the user's preferences and interests are more effectively obtained, and the user is provided with search results that more satisfy the user's needs, thereby improving the user experience.
- the time at which the behavior data is generated and the current time are dates
- the calculation module 200 is configured to calculate a time difference between the date at which the behavior data is generated and the current date, and calculate the aging according to the time difference and the time decay function.
- Sex factor the time decay function is an exponential function or a power function.
- the time and current time at which the behavior data is generated belong to the ith day and the jth day in the preset time period, wherein the preset time period has N days, and i and j are less than or equal to A positive integer of N, wherein the calculation module 200 is further configured to calculate a date difference between the ith day and the current jth day of the behavior data, and generate a periodicity factor according to the date difference.
- the periodic factor is generated by the following formula: Where di is the ith day when the behavior data is generated and dj is the current jth day.
- the processing module 300 is configured to adjust the behavior data according to the time-effect factor and the periodic factor, and use the product of the time-effect factor and the behavior data as the first behavior data, and according to the periodic factor. , behavior data and first behavior data generation second line For the data.
- the processing module 300 is further configured to generate third behavior data according to the periodicity factor and the behavior data, and generate the second behavior data according to the first behavior data and the third behavior data. Wherein the first behavior data and the third behavior data are added or multiplied to generate second behavior data.
- the device for processing user behavior data in the embodiment of the present application firstly performs time-sensitive processing on the behavior data, and then periodically processes the behavior data after the time-sensitive processing, thereby passing the machine learning method according to the behavior data after the periodic processing.
- first and second are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated.
- features defining “first” or “second” may include at least one of the features, either explicitly or implicitly.
- the meaning of "a plurality” is at least two, such as two, three, etc., unless specifically defined otherwise.
- portions of the application can be implemented in hardware, software, firmware, or a combination thereof.
- multiple steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system.
- a suitable instruction execution system For example, if implemented in hardware, as in another embodiment, it can be implemented by any one or combination of the following techniques well known in the art: having logic gates for implementing logic functions on data signals. Discrete logic circuits, application specific integrated circuits with suitable combinational logic gates, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.
- each functional unit in each embodiment of the present application may be integrated into one processing module, or each unit may exist physically separately, or two or more units may be integrated into one module.
- the above integrated modules can be implemented in the form of hardware or in the form of software functional modules.
- the integrated modules, if implemented in the form of software functional modules and sold or used as stand-alone products, may also be stored in a computer readable storage medium.
- the above mentioned storage medium may be a read only memory, a magnetic disk or an optical disk or the like. While the embodiments of the present application have been shown and described above, it is understood that the above-described embodiments are illustrative and are not to be construed as limiting the scope of the present application. The embodiments are subject to variations, modifications, substitutions and variations.
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Computer Hardware Design (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Physics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Operations Research (AREA)
- Probability & Statistics with Applications (AREA)
- Algebra (AREA)
- Software Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
Abstract
一种用户行为数据的处理方法和装置,其中,处理方法包括以下步骤:获取用户的行为数据,并获取产生行为数据的时间(S110);根据产生行为数据的时间和当前时间计算行为数据对应的时效性因子和/或周期性因子(S120);根据时效性因子和/或周期性因子对行为数据进行调整(S130)。实施例的处理方法,可以提升用户行为数据的时效性,能够更加有效地获取到用户的偏好和兴趣,为用户提供更加满足用户需求的搜索结果,提升了用户的使用体验。
Description
本申请涉及互联网技术领域,尤其涉及一种用户行为数据的处理方法和装置。
搜索引擎经常用机器学习方法通过对历史日志数据的学习来优化多个搜索结果的排序结果,为了能够取得更理想的搜索结果,往往会使用过去两周、一个月甚至更长时间的历史数据来预测用户对搜索结果的偏好。
然而,目前存在的问题是,搜索引擎在通过历史日志数据优化搜索结果的排序结果时通常是平等对待所有的历史数据,例如,一个10天前的行为日志和一个1天前的历史数据,搜索引擎在根据行为日志预测用户偏好时并不区分历史数据的重要程度,将所有的历史数据按照同样的重要程度用机器学习方法预测用户的偏好,导致搜索引擎无法有效地捕捉用户的兴趣变化,不能满足用户的需求。
发明内容
本申请的目的旨在至少在一定程度上解决上述的技术问题之一。
为此,本申请的第一个目的在于提出一种用户行为数据的处理方法,该处理方法可以提升用户行为数据的时效性,能够更加有效地获取到用户的偏好和兴趣,为用户提供更加满足用户需求的搜索结果,提升了用户的使用体验。
本申请的第二个目的在于提出了一种用户行为数据的处理装置。
为达上述目的,根据本申请第一方面实施例提出的一种用户行为数据的处理方法,包括以下步骤:获取用户的行为数据,并获取产生所述行为数据的时间;根据所述产生所述行为数据的时间和当前时间计算所述行为
数据对应的时效性因子和/或周期性因子;根据所述时效性因子和/或周期性因子对所述行为数据进行调整。
本申请实施例的处理方法,结合时间衰减和周期性相似性为用户的行为数据赋予不同的重要程度,利用不同重要程度的行为数据通过机器学习方法优化多个搜索结果的排序结果,从而提升了用户行为数据的时效性,能够更加有效地获取到用户的偏好和兴趣,为用户提供更加满足用户需求的搜索结果,提升了用户的使用体验。
为达上述目的,根据本申请的第二方面实施例提出的一种用户行为数据的处理装置,包括:获取模块,用于获取用户的行为数据,并获取产生所述行为数据的时间;计算模块,用于根据所述产生所述行为数据的时间和当前时间计算所述行为数据对应的时效性因子和/或周期性因子;处理模块,用于根据所述时效性因子和/或周期性因子对所述行为数据进行调整。
本申请实施例的处理装置,结合时间衰减和周期性相似性为用户的行为数据赋予不同的重要程度,利用不同重要程度的行为数据通过机器学习方法优化多个搜索结果的排序结果,从而提升了用户行为数据的时效性,能够更加有效地获取到用户的偏好和兴趣,为用户提供更加满足用户需求的搜索结果,提升了用户的使用体验。
本申请附加的方面和优点将在下面的描述中部分给出,部分将从下面的描述中变得明显,或通过本申请的实践了解到。
本申请的上述和/或附加的方面和优点从结合下面附图对实施例的描述中将变得明显和容易理解,其中:
图1是根据本申请一个实施例的用户行为数据的处理方法的流程图;
图2是本发明一个具体实施例的用户行为数据的处理方法的流程图;
图3是本发明另一个具体实施例的用户行为数据的处理方法的流程图;以及
图4是根据本申请一个实施例的用户行为数据的处理装置的结构示意图。
下面详细描述本申请的实施例,所述实施例的示例在附图中示出,其中自始至终相同或类似的标号表示相同或类似的元件或具有相同或类似功能的元件。下面通过参考附图描述的实施例是示例性的,旨在用于解释本申请,而不能理解为对本申请的限制。
下面参考附图描述本申请实施例的用户行为数据的处理方法和装置。
图1是根据本申请一个实施例的用户行为数据的处理方法的流程图。
如图1所示,该处理方法包括:
S110,获取用户的行为数据,并获取产生行为数据的时间。
例如,行为数据的格式一般如下表所示,其中,features为行为数据中的特征,label为用户的点击行为(用^分为两部分,第一部分为1表示用户点击,第二部分为1表示用户没有点击),date为产生点击行为的时间。
| features | label | date |
| city_1,age_2,age_city_4 | 0^1 | 2015-11-24 |
| city_6,age_20,age_city_3 | 1^0 | 2015-11-20 |
| city_199,age_10,age_city_1 | 0^1 | 2015-11-02 |
S120,根据产生行为数据的时间和当前时间计算行为数据对应的时效性因子和/或周期性因子。
例如,用户A在过去一个月里多次访问手机类的页面,然而,在过去的一周内用户浏览手机类页面和在一个月前用户浏览手机类页面的影响
是不同的,显然离现在越近时间点的用户行为的影响更大,更能符合用户的搜索需求。因此,在使用历史数据学习时,可以按照时间远近给历史数据赋予不同的重要程度。
再例如,有些用户行为是有周期性的,比如上班族用户B经常在周末的时候在网上购买食材,因为用户B周末的时候才有时间做饭。那么虽然周一离上个周末的时间上很接近,但是用户B的行为差异仍然很大。因此,在使用历史数据学习时,需要考虑周期性的影响,根据历史数据在某个维度上与当前周期性的相近程度来给历史数据赋予不同的重要程度。
具体地,计算产生行为数据的时间与当前时间的时间差,根据特定的函数计算时效性因子Z。根据产生行为数据的时间与当前时间,通过特定的函数计算出周期性因子C。
在本发明的一个实施例中,在计算时效性因子Z和周期性因子C之前,还可以先将行为数据输入至数据切分装置进行切分,将行为数据切分为多个小的行为数据块。其中,数据切分装置根据行为数据的大小,以及处理行为数据的每台计算机器可以计算的数据量判断将行为数据切分成多少份。由于数据切分装置对行为数据的切分本身是在分布式环境下执行的,因此可以充分利用多个计算机器的资源来加速切分动作。数据切分装置将行为数据切分好后,将每个小的行为数据块分发至每台计算机器上,多台计算机器可以同时并行处理行为数据块,其中任意两台计算机器获取到的行为数据块不会重叠。
S130,根据时效性因子和/或周期性因子对行为数据进行调整。
具体地,根据时效性因子Z与用户的点击行为label计算得到新的点击行为label1,即label1=Z*label。进而,根据周期性因子C与新的点击行为label1计算得到新的点击行为label2。也就是说,首先对用户的行为数据施加时效性和周期性的影响,然后利用得到的调整后的行为数据通过机器学习方法优化多个搜索结果的排序结果。
应当理解的是,在对用户的行为数据进行调整时,可以仅根据时效性因子Z对用户的行为数据进行调整,或者仅根据周期性因子C对用户的行
为数据进行调整,或者还可以同时根据时效性因子Z和周期性因子C对用户的行为数据进行调整,即先根据时效性因子Z对用户的行为数据进行调整,再根据周期性因子C对用户的行为数据进行调整。
本申请实施例中的用户行为数据的处理方法,结合时间衰减和周期性相似性为用户的行为数据赋予不同的重要程度,利用不同重要程度的行为数据通过机器学习方法优化多个搜索结果的排序结果,从而提升了用户行为数据的时效性,能够更加有效地获取到用户的偏好和兴趣,为用户提供更加满足用户需求的搜索结果,提升了用户的使用体验。
为了使得本申请中对用户行为数据的处理方法的描述更加的具体化,使得本领域的技术人员更加清楚的了解本申请,下面以同时根据时效性因子Z和周期性因子C对用户的行为数据进行调整的具体示例进行说明。
图2是本发明一个具体实施例的用户行为数据的处理方法的流程图,图3是本发明另一个具体实施例的用户行为数据的处理方法的流程图,如图2和3所示,用户行为数据的处理方法包括:
S210,计算产生行为数据的日期与当前日期之间的时间差。其中,产生行为数据的时间和当前时间为日期。
S220,根据时间差和时间衰减函数计算时效性因子。其中,时间衰减函数为指数函数或幂函数。
S230,将时效性因子与行为数据的乘积作为第一行为数据。
S240,计算产生行为数据的第i日和当前第j日的日期差。其中,产生行为数据的时间和当前时间属于预设时间周期中的第i日和第j日,其中预设时间周期具有N日,i和j为小于或等于N的正整数。
S250,根据日期差生成周期性因子。其中,通过以下公式生成周期性因子:
其中,di为产生行为数据的第i日,dj为当前第j日。
S260,根据周期性因子、行为数据与第一行为数据生成第二行为数据。
在本发明的一个实施例中,如图3所示,根据周期性因子、行为数据与第一行为数据生成第二行为数据具体包括:
S261,根据周期性因子和行为数据生成第三行为数据。
S262,根据第一行为数据和第三行为数据生成第二行为数据。其中,将第一行为数据和第三行为数据相加或相乘以生成第二行为数据。
具体而言,首先,输入用户行为数据中的用户的点击行为label和产生点击行为的时间date,并按照时间衰减规律来给用户的行为数据赋予权重。应当理解的时,在对用户的行为数据进行切分后,由于只需要输入用户的点击行为label和产生点击行为的时间date,因此可以大大降低每台计算机器的数据处理量。设t0是当前日期,t是产生行为数据的日期,因此,t0-t可以衡量用户的行为数据离当前日期的远近程度,也就是距离。如果时间衰减函数f(x)是一个递减函数,即f(x)在[0,+∞)上是一个递减函数,那么f(t0-t)就构成一个按时间衰减给行为数据进行加权的方法。例如,一天前的行为数据权重是f(1),三天前的行为数据权重是f(3),以此类推。由于f(x)为递减函数,f(1)>f(3),因而离现在时间间隔越大的行为数据的权重越小。其中,时间衰减函数f(x)可以时指数函数f(x)=ax,0<α<1,还可以是幂函数f(x)=x-α,α>0等。
应当理解的是,在按照时间衰减规律给用户的行为数据赋予权重时,当前时间和产生行为数据的时间可以以天为单位,即当前日期和产生行为数据的日期,也可以以小时或者分钟等为单位,本实施例中仅以以天为单位进行说明,以小时或分钟为单位计算行为数据的权重和以天为单位计算行为数据的权重的方式相同,为了避免冗余,此处不在复赘。
进而,计算产生行为数据的日期与当前日期之间的差值得到时间差,即将当前日期减去行为数据中date的日期以得到时间差,然后选择指数函数或者幂函数计算得到时效性因子Z,例如时效性因子Z为指数函数f(x)=0.98x,然后计算时效性处理后的用户的点击行为label1,label1等于用户的点击行为label乘以时效性因子Z,即label1=Z*label。
然后,输入对用户的时效性处理后的用户的点击行为label1、用户的点
击行为label以及产生点击行为的时间date,并根据时间差通过公式得到周期性因子C。例如,假设产生行为数据的是一周中的第di天,以一周中的七天为例,以数字1到7来记录周一到周日,因此当前第dj天时行为数据的权重就应该是其中,N等于7。
应当理解的是,这里仅以一周中的第几天来描述周期性信息,任何其他周期性的单位都可以用类似方法衡量,例如,一天中的24小时,节气的变更,温度的升降等周期性信息。
进而,根据选择的叠加关系计算得到label2。其中,叠加关系可以为加或者乘等,如果叠加关系是加,则:
如果叠加关系是乘,则:
然后,然后利用得到的调整后的行为数据label2通过机器学习方法优化多个搜索结果的排序结果。
本申请实施例中的用户行为数据的处理方法,先对行为数据进行时效性的处理,再对时效性处理后的行为数据进行周期性处理,从而根据周期性处理后的行为数据通过机器学习方法优化多个搜索结果的排序结果以捕捉用户的偏好,从而提升了用户行为数据的时效性,能够更加有效地获取到用户的偏好和兴趣,为用户提供更加满足用户需求的搜索结果,提升了用户的使用体验。
为了实现上述实施例,本申请还提出一种用户行为数据的处理装置。
图4是根据本申请一个实施例的用户行为数据的处理装置的结构示意图。
如图4所示,该处理装置包括:获取模块100、计算模块200和处理模块300。
其中,获取模块100用于获取用户的行为数据,并获取产生行为数据
的时间。
计算模块200用于根据产生行为数据的时间和当前时间计算行为数据对应的时效性因子和/或周期性因子。
处理模块300用于根据时效性因子和/或周期性因子对行为数据进行调整。
需要说明的是,前述对用户行为数据的处理方法的解释说明也适用于该实施例的用户行为数据的处理装置,其实现原理类似,此处不再赘述。
本申请实施例中的用户行为数据的处理装置,结合时间衰减和周期性相似性为用户的行为数据赋予不同的重要程度,利用不同重要程度的行为数据通过机器学习方法优化多个搜索结果的排序结果,从而提升了用户行为数据的时效性,能够更加有效地获取到用户的偏好和兴趣,为用户提供更加满足用户需求的搜索结果,提升了用户的使用体验。
为了使得本申请中对用户行为数据的处理装置的描述更加的具体化,使得本领域的技术人员更加清楚的了解本申请,下面以同时根据时效性因子Z和周期性因子C对用户的行为数据进行调整的具体示例进行说明。
在本申请的一个实施例中,产生行为数据的时间和当前时间为日期,其中,计算模块200用于计算产生行为数据的日期与当前日期之间的时间差,并根据时间差和时间衰减函数计算时效性因子。其中,时间衰减函数为指数函数或幂函数。
在本申请的另一个实施例中,产生行为数据的时间和当前时间属于预设时间周期中的第i日和第j日,其中,预设时间周期具有N日,i和j为小于或等于N的正整数,其中,计算模块200还用于计算产生行为数据的第i日和当前第j日的日期差,并根据日期差生成周期性因子。其中,通过以下公式生成周期性因子:其中,di为产生行为数据的第i日,dj为当前第j日。
在本申请的另一个实施例中,处理模块300用于根据时效性因子和周期性因子对行为数据进行调整,并将时效性因子与行为数据的乘积作为第一行为数据,以及根据周期性因子、行为数据与第一行为数据生成第二行
为数据。
在本申请的另一个实施例中,处理模块300还用于根据周期性因子和行为数据生成第三行为数据,并根据第一行为数据和第三行为数据生成第二行为数据。其中,将第一行为数据和第三行为数据相加或相乘以生成第二行为数据。
需要说明的是,前述对用户行为数据的处理方法的解释说明也适用于该实施例的用户行为数据的处理装置,其实现原理类似,此处不再赘述。
本申请实施例中的用户行为数据的处理装置,先对行为数据进行时效性的处理,再对时效性处理后的行为数据进行周期性处理,从而根据周期性处理后的行为数据通过机器学习方法优化多个搜索结果的排序结果以捕捉用户的偏好,从而提升了用户行为数据的时效性,能够更加有效地获取到用户的偏好和兴趣,为用户提供更加满足用户需求的搜索结果,提升了用户的使用体验。
在本申请的描述中,需要理解的是,术语“第一”、“第二”仅用于描述目的,而不能理解为指示或暗示相对重要性或者隐含指明所指示的技术特征的数量。由此,限定有“第一”、“第二”的特征可以明示或者隐含地包括至少一个该特征。在本申请的描述中,“多个”的含义是至少两个,例如两个,三个等,除非另有明确具体的限定。
在本说明书的描述中,参考术语“一个实施例”、“一些实施例”、“示例”、“具体示例”、或“一些示例”等的描述意指结合该实施例或示例描述的具体特征、结构、材料或者特点包含于本申请的至少一个实施例或示例中。在本说明书中,对上述术语的示意性表述不必须针对的是相同的实施例或示例。而且,描述的具体特征、结构、材料或者特点可以在任一个或多个实施例或示例中以合适的方式结合。此外,在不相互矛盾的情况下,本领域的技术人员可以将本说明书中描述的不同实施例或示例以及不同实施例或示例的特征进行结合和组合。
流程图中或在此以其他方式描述的任何过程或方法描述可以被理解为,表示包括一个或更多个用于实现特定逻辑功能或过程的步骤的可执行指令的代码的模块、片段或部分,并且本申请的优选实施方式的范围包括另外的实现,其中可以不按所示出或讨论的顺序,包括根据所涉及的功能按
基本同时的方式或按相反的顺序,来执行功能,这应被本申请的实施例所属技术领域的技术人员所理解。
应当理解,本申请的各部分可以用硬件、软件、固件或它们的组合来实现。在上述实施方式中,多个步骤或方法可以用存储在存储器中且由合适的指令执行系统执行的软件或固件来实现。例如,如果用硬件来实现,和在另一实施方式中一样,可用本领域公知的下列技术中的任一项或他们的组合来实现:具有用于对数据信号实现逻辑功能的逻辑门电路的离散逻辑电路,具有合适的组合逻辑门电路的专用集成电路,可编程门阵列(PGA),现场可编程门阵列(FPGA)等。
本技术领域的普通技术人员可以理解实现上述实施例方法携带的全部或部分步骤是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,该程序在执行时,包括方法实施例的步骤之一或其组合。
此外,在本申请各个实施例中的各功能单元可以集成在一个处理模块中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个模块中。上述集成的模块既可以采用硬件的形式实现,也可以采用软件功能模块的形式实现。所述集成的模块如果以软件功能模块的形式实现并作为独立的产品销售或使用时,也可以存储在一个计算机可读取存储介质中。
上述提到的存储介质可以是只读存储器,磁盘或光盘等。尽管上面已经示出和描述了本申请的实施例,可以理解的是,上述实施例是示例性的,不能理解为对本申请的限制,本领域的普通技术人员在本申请的范围内可以对上述实施例进行变化、修改、替换和变型。
Claims (16)
- 一种用户行为数据的处理方法,其特征在于,包括以下步骤:获取用户的行为数据,并获取产生所述行为数据的时间;根据所述产生所述行为数据的时间和当前时间计算所述行为数据对应的时效性因子和/或周期性因子;根据所述时效性因子和/或周期性因子对所述行为数据进行调整。
- 如权利要求1所述的用户行为数据的处理方法,其特征在于,所述根据所述时效性因子和所述周期性因子对所述行为数据进行调整;将所述时效性因子与所述行为数据的乘积作为第一行为数据;根据所述周期性因子、所述行为数据与所述第一行为数据生成第二行为数据。
- 如权利要求1所述的用户行为数据的处理方法,其特征在于,所述产生所述行为数据的时间和所述当前时间为日期,其中,所述根据所述产生所述行为数据的时间和当前时间计算所述行为数据对应的时效性因子具体包括:计算所述产生所述行为数据的日期与所述当前日期之间的时间差;根据所述时间差和时间衰减函数计算所述时效性因子。
- 如权利要求3所述的用户行为数据的处理方法,其特征在于,所述时间衰减函数为指数函数或幂函数。
- 如权利要求1所述的用户行为数据的处理方法,其特征在于,所述产生所述行为数据的时间和所述当前时间属于预设时间周期中的第i日和第j日,其中,所述预设时间周期具有N日,i和j为小于或等于N的正整数,其中,所述根据所述产生所述行为数据的时间和当前时间计算所述行为数据对应的周期性因子具体包括:计算产生所述行为数据的所述第i日和当前第j日的日期差;根据所述日期差生成所述周期性因子。
- 如权利要求2所述的用户行为数据的处理方法,其特征在于,所述根据所述周期性因子、所述行为数据与所述第一行为数据生成第二行为数据具体包括:根据所述周期性因子和所述行为数据生成第三行为数据;根据所述第一行为数据和所述第三行为数据生成所述第二行为数据。
- 如权利要求7所述的用户行为数据的处理方法,其特征在于,将所述第一行为数据和所述第三行为数据相加或相乘以生成所述第二行为数据。
- 一种用户行为数据的处理装置,其特征在于,包括:获取模块,用于获取用户的行为数据,并获取产生所述行为数据的时间;计算模块,用于根据所述产生所述行为数据的时间和当前时间计算所述行为数据对应的时效性因子和/或周期性因子;处理模块,用于根据所述时效性因子和/或周期性因子对所述行为数据进行调整。
- 如权利要求9所述的用户行为数据的处理装置,其特征在于,所述处理模块用于:根据所述时效性因子和所述周期性因子对所述行为数据进行调整,并将所述时效性因子与所述行为数据的乘积作为第一行为数据,以及根据所述周期性因子、所述行为数据与所述第一行为数据生成第二行为数据。
- 如权利要求9所述的用户行为数据的处理装置,其特征在于,所述产生所述行为数据的时间和所述当前时间为日期,其中,所述计算模块用于:计算所述产生所述行为数据的日期与所述当前日期之间的时间差,并根据所述时间差和时间衰减函数计算所述时效性因子。
- 如权利要求11所述的用户行为数据的处理装置,其特征在于,所述时间衰减函数为指数函数或幂函数。
- 如权利要求1所述的用户行为数据的处理装置,其特征在于,所述产生所述行为数据的时间和所述当前时间属于预设时间周期中的第i日和第j日,其中,所述预设时间周期具有N日,i和j为小于或等于N的正整数,其中,所述计算模块用于:计算产生所述行为数据的所述第i日和当前第j日的日期差,并根据所述日期差生成所述周期性因子。
- 如权利要求10所述的用户行为数据的处理装置,其特征在于,所述处理模块还用于:根据所述周期性因子和所述行为数据生成第三行为数据,并根据所述第一行为数据和所述第三行为数据生成所述第二行为数据。
- 如权利要求15所述的用户行为数据的处理装置,其特征在于,将所述第一行为数据和所述第三行为数据相加或相乘以生成所述第二行为数据。
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2018538558A JP7085486B2 (ja) | 2016-01-25 | 2017-01-16 | ユーザ挙動データの処理方法及び装置 |
| US16/045,625 US20180330001A1 (en) | 2016-01-25 | 2018-07-25 | Method and apparatus for processing user behavior data |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610048385.8A CN106997360A (zh) | 2016-01-25 | 2016-01-25 | 用户行为数据的处理方法和装置 |
| CN201610048385.8 | 2016-01-25 |
Related Child Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| US16/045,625 Continuation US20180330001A1 (en) | 2016-01-25 | 2018-07-25 | Method and apparatus for processing user behavior data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2017128973A1 true WO2017128973A1 (zh) | 2017-08-03 |
Family
ID=59397359
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2017/071241 Ceased WO2017128973A1 (zh) | 2016-01-25 | 2017-01-16 | 用户行为数据的处理方法和装置 |
Country Status (5)
| Country | Link |
|---|---|
| US (1) | US20180330001A1 (zh) |
| JP (1) | JP7085486B2 (zh) |
| CN (1) | CN106997360A (zh) |
| TW (1) | TW201732642A (zh) |
| WO (1) | WO2017128973A1 (zh) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108241970B (zh) * | 2018-01-09 | 2021-05-18 | 重庆金窝窝网络科技有限公司 | 基于区块链的挖矿方法、装置及计算机可读存储介质 |
| US10824624B2 (en) * | 2018-07-12 | 2020-11-03 | Bank Of America Corporation | System for analyzing, optimizing, and remediating a proposed data query prior to query implementation |
| CN112883257B (zh) * | 2021-01-11 | 2024-01-05 | 北京达佳互联信息技术有限公司 | 行为序列数据处理方法、装置、电子设备及存储介质 |
| CN113205217B (zh) * | 2021-05-07 | 2022-07-15 | 上海一谈网络科技有限公司 | 数据处理方法、装置、设备及存储介质 |
| CN113868536B (zh) * | 2021-10-13 | 2024-09-13 | 平安银行股份有限公司 | 一种信息推荐方法、装置、设备和存储介质 |
| CN117033756B (zh) * | 2022-10-18 | 2025-08-08 | 腾讯科技(深圳)有限公司 | 资源内容推送方法及相关设备 |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090089233A1 (en) * | 2007-09-28 | 2009-04-02 | Rockwell Automation Technologies, Inc. | Correlation of non-times series events in industrial systems |
| CN102075352A (zh) * | 2010-12-17 | 2011-05-25 | 北京邮电大学 | 一种网络用户行为预测的方法和装置 |
| CN103164804A (zh) * | 2011-12-16 | 2013-06-19 | 阿里巴巴集团控股有限公司 | 一种个性化的信息推送方法及装置 |
| CN104216884A (zh) * | 2013-05-29 | 2014-12-17 | 酷盛(天津)科技有限公司 | 一种基于时间衰减的协同过滤系统及方法 |
Family Cites Families (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7364086B2 (en) * | 2003-06-16 | 2008-04-29 | Ewinwin, Inc. | Dynamic discount card tied to price curves and group discounts |
| US7505964B2 (en) * | 2003-09-12 | 2009-03-17 | Google Inc. | Methods and systems for improving a search ranking using related queries |
| US8954420B1 (en) * | 2003-12-31 | 2015-02-10 | Google Inc. | Methods and systems for improving a search ranking using article information |
| US8255413B2 (en) * | 2004-08-19 | 2012-08-28 | Carhamm Ltd., Llc | Method and apparatus for responding to request for information-personalization |
| US20070061195A1 (en) * | 2005-09-13 | 2007-03-15 | Yahoo! Inc. | Framework for selecting and delivering advertisements over a network based on combined short-term and long-term user behavioral interests |
| US20070260597A1 (en) * | 2006-05-02 | 2007-11-08 | Mark Cramer | Dynamic search engine results employing user behavior |
| US9009146B1 (en) * | 2009-04-08 | 2015-04-14 | Google Inc. | Ranking search results based on similar queries |
| JP5591608B2 (ja) * | 2009-09-15 | 2014-09-17 | 株式会社Nttドコモ | 情報提供システム、情報提供方法、及び情報提供プログラム |
| US9081857B1 (en) * | 2009-09-21 | 2015-07-14 | A9.Com, Inc. | Freshness and seasonality-based content determinations |
| CN102339296A (zh) * | 2010-07-26 | 2012-02-01 | 阿里巴巴集团控股有限公司 | 一种查询结果的排序方法和装置 |
| CN102346894B (zh) * | 2010-08-03 | 2017-03-01 | 阿里巴巴集团控股有限公司 | 推荐信息的输出方法、系统及服务器 |
| JP5192518B2 (ja) * | 2010-08-04 | 2013-05-08 | 三菱電機インフォメーションシステムズ株式会社 | 文書検索システム、文書検索方法および文書検索プログラム |
| US20120143790A1 (en) * | 2010-12-01 | 2012-06-07 | Microsoft Corporation | Relevance of search results determined from user clicks and post-click user behavior obtained from click logs |
| CN102890689B (zh) * | 2011-07-22 | 2017-06-06 | 北京百度网讯科技有限公司 | 一种用户兴趣模型的建立方法及系统 |
| CN102956009B (zh) * | 2011-08-16 | 2017-03-01 | 阿里巴巴集团控股有限公司 | 一种基于用户行为的电子商务信息推荐方法与装置 |
| US9355095B2 (en) * | 2011-12-30 | 2016-05-31 | Microsoft Technology Licensing, Llc | Click noise characterization model |
| CN103310343A (zh) * | 2012-03-15 | 2013-09-18 | 阿里巴巴集团控股有限公司 | 商品信息发布方法和装置 |
| CN103714084B (zh) * | 2012-10-08 | 2018-04-03 | 腾讯科技(深圳)有限公司 | 推荐信息的方法和装置 |
| US20140280120A1 (en) * | 2013-03-15 | 2014-09-18 | Highspot, Inc. | Interest graph-powered browsing |
| US9946794B2 (en) * | 2014-10-30 | 2018-04-17 | Samsung Electronics Co., Ltd. | Accessing special purpose search systems |
| US10423681B2 (en) * | 2015-09-11 | 2019-09-24 | Walmart Apollo, Llc | System for hybrid incremental approach to query processing and method therefor |
-
2016
- 2016-01-25 CN CN201610048385.8A patent/CN106997360A/zh active Pending
-
2017
- 2017-01-16 JP JP2018538558A patent/JP7085486B2/ja active Active
- 2017-01-16 WO PCT/CN2017/071241 patent/WO2017128973A1/zh not_active Ceased
- 2017-01-23 TW TW106102453A patent/TW201732642A/zh unknown
-
2018
- 2018-07-25 US US16/045,625 patent/US20180330001A1/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090089233A1 (en) * | 2007-09-28 | 2009-04-02 | Rockwell Automation Technologies, Inc. | Correlation of non-times series events in industrial systems |
| CN102075352A (zh) * | 2010-12-17 | 2011-05-25 | 北京邮电大学 | 一种网络用户行为预测的方法和装置 |
| CN103164804A (zh) * | 2011-12-16 | 2013-06-19 | 阿里巴巴集团控股有限公司 | 一种个性化的信息推送方法及装置 |
| CN104216884A (zh) * | 2013-05-29 | 2014-12-17 | 酷盛(天津)科技有限公司 | 一种基于时间衰减的协同过滤系统及方法 |
Also Published As
| Publication number | Publication date |
|---|---|
| US20180330001A1 (en) | 2018-11-15 |
| CN106997360A (zh) | 2017-08-01 |
| JP7085486B2 (ja) | 2022-06-16 |
| TW201732642A (zh) | 2017-09-16 |
| JP2019503014A (ja) | 2019-01-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US12436512B2 (en) | Determining causal models for controlling environments | |
| WO2017128973A1 (zh) | 用户行为数据的处理方法和装置 | |
| US11170027B2 (en) | Error factor and uniqueness level for anonymized datasets | |
| CN109903086B (zh) | 一种相似人群扩展方法、装置及电子设备 | |
| US20150242447A1 (en) | Identifying effective crowdsource contributors and high quality contributions | |
| BR102013016747A2 (pt) | Sistema e método para fornecer conteúdo eletrônico | |
| CN102880688A (zh) | 一种用于对网页进行评估的方法、装置和设备 | |
| CN104123377A (zh) | 一种微博话题热度预测系统及方法 | |
| HK1239890A1 (zh) | 用戶行為數據的處理方法和裝置 | |
| HK1239890A (zh) | 用户行为数据的处理方法和装置 | |
| US12468781B2 (en) | System and method for reduction of data transmission by information control with reinforced learning | |
| US20230126932A1 (en) | Recommended audience size | |
| US20250265466A1 (en) | Adaptive management of data source unavailability for an inference model | |
| CN121743432A (zh) | 用于模型确定的方法、装置、设备和存储介质 |
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: 17743588 Country of ref document: EP Kind code of ref document: A1 |
|
| ENP | Entry into the national phase |
Ref document number: 2018538558 Country of ref document: JP Kind code of ref document: A |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 17743588 Country of ref document: EP Kind code of ref document: A1 |