CN112052380B - Marketing recommendation method and system for smart terminals and smart home devices - Google Patents
Marketing recommendation method and system for smart terminals and smart home devices Download PDFInfo
- Publication number
- CN112052380B CN112052380B CN202010067606.2A CN202010067606A CN112052380B CN 112052380 B CN112052380 B CN 112052380B CN 202010067606 A CN202010067606 A CN 202010067606A CN 112052380 B CN112052380 B CN 112052380B
- Authority
- CN
- China
- Prior art keywords
- user
- pool
- recommendation
- host
- type
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- 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
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/231—Hierarchical techniques, i.e. dividing or merging pattern sets so as to obtain a dendrogram
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Probability & Statistics with Applications (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
Description
技术领域Technical Field
本发明涉及一种智能终端、智能家居设备的营销推荐方法及系统,属于智能家居技术领域。The present invention relates to a marketing recommendation method and system for a smart terminal and smart home equipment, belonging to the technical field of smart homes.
背景技术Background Art
随着大数据时代到来,各行各业数据采集和存储设备不断健全,用户的数据量无论从时间维度还是空间维度都在迅速增加。智能家居行业更是数据产量大户,这些数据可以实现智能家居设备的精准营销,为用户推荐适合用户的可购买的智能家居产品。With the advent of the big data era, data collection and storage equipment in various industries are constantly improving, and the amount of user data is increasing rapidly in both time and space dimensions. The smart home industry is a big producer of data. These data can achieve precise marketing of smart home devices and recommend smart home products that are suitable for users.
目前,现有技术中用于营销的算法有以下几种:At present, there are several algorithms used for marketing in the existing technology:
1.聚类算法。作为无监督学习的聚类算法,主要分为以下几类:基于划分的聚类算法、基于层次的聚类算法以及基于密度的聚类算法等。例如:DBSCAN聚类算法,该算法可以对任意形状的稠密数据集进行聚类,抗异常点干扰能力强;然而当空间聚类的密度不均匀、聚类间距差相差很大时,聚类质量较差,导致推荐结果的准确度下降,用户的体检效果不佳。1. Clustering algorithm. As an unsupervised learning clustering algorithm, it is mainly divided into the following categories: partition-based clustering algorithm, hierarchical-based clustering algorithm, and density-based clustering algorithm. For example: DBSCAN clustering algorithm, which can cluster dense data sets of any shape and has strong resistance to abnormal point interference; however, when the density of spatial clustering is uneven and the distance between clusters varies greatly, the clustering quality is poor, resulting in a decrease in the accuracy of the recommendation results and poor physical examination results for users.
2.协同过滤算法。协同过滤算法作为推荐系统的流行算法,被广泛应用于各大电商平台。协同过滤算法分为基于用户和基于物品两类,基于物品的协同过滤算法很难提供令用户信服的推荐解释,会使用户体验变差;基于用户的协同过滤算法适用于用户较少的场合,如果用户很多,计算用户相似度矩阵代价很大,推荐效率低,用户的体验效果不佳。2. Collaborative filtering algorithm. As a popular algorithm for recommendation systems, collaborative filtering algorithms are widely used on major e-commerce platforms. Collaborative filtering algorithms are divided into two categories: user-based and item-based. Item-based collaborative filtering algorithms are difficult to provide users with convincing recommendation explanations, which will deteriorate the user experience; user-based collaborative filtering algorithms are suitable for occasions with fewer users. If there are many users, the cost of calculating the user similarity matrix is very high, the recommendation efficiency is low, and the user experience is not good.
3.时间衰减算法。电商行业在给用户推荐商品时,会分析用户对于平台商品的兴趣偏好度,同时这个兴趣偏好度也会随着时间的流逝而发生变化。时间衰减算法考虑了用户兴趣偏好与时间远近有关,然而由于该算法侧重于考虑用户近期的兴趣偏好而使得整体推荐并不准确,用户的体检效果不佳。3. Time decay algorithm. When the e-commerce industry recommends products to users, it will analyze the user's interest preference for platform products, and this interest preference will also change over time. The time decay algorithm takes into account that the user's interest preference is related to the distance in time. However, because the algorithm focuses on considering the user's recent interest preference, the overall recommendation is not accurate, and the user's physical examination effect is not good.
综上,现有的算法推荐不准确、效率低,用户的体验效果不佳。In summary, existing algorithm recommendations are inaccurate and inefficient, resulting in poor user experience.
发明内容Summary of the invention
本发明的目的在于提供一种智能家居设备的营销推荐方法,用以解决现有营销推荐方法推荐准确度低、效率低,用户体验效果不佳的问题;同时还提供一种智能家居设备的营销推荐系统,用以解决现有推荐系统推荐准确度低、效率低,用户体验效果不佳的问题;同时还提供一种智能终端,用以解决现有智能终端推荐准确度低、效率低,用户体验效果不佳的问题。The purpose of the present invention is to provide a marketing recommendation method for smart home devices to solve the problems of low recommendation accuracy, low efficiency and poor user experience of existing marketing recommendation methods; at the same time, a marketing recommendation system for smart home devices is provided to solve the problems of low recommendation accuracy, low efficiency and poor user experience of existing recommendation systems; at the same time, a smart terminal is provided to solve the problems of low recommendation accuracy, low efficiency and poor user experience of existing smart terminals.
为实现上述目的,本发明提出一种智能家居设备的营销推荐方法,包括以下步骤:To achieve the above object, the present invention proposes a marketing recommendation method for smart home devices, comprising the following steps:
根据每台主机所绑定的设备清单,采用聚类算法对所有主机进行分类;Based on the device list bound to each host, a clustering algorithm is used to classify all hosts;
统计每类主机下的用户,构成每类主机的用户池;Count the users of each type of host and form a user pool for each type of host;
对于某一类主机的用户池,根据用户近期的设备点击次数,采用时间衰减算法得到该用户池中每个用户的推荐向量;For a user pool of a certain type of host, a time decay algorithm is used to obtain the recommendation vector of each user in the user pool based on the number of recent device clicks by the user;
根据该用户池中每个用户的推荐向量,采用聚类算法对该用户池中的用户进行分类;According to the recommendation vector of each user in the user pool, a clustering algorithm is used to classify the users in the user pool;
统计该用户池中每类用户所在主机绑定的设备清单,构成该用户池中每类用户的设备池;Count the device list bound to the host of each type of user in the user pool to form a device pool for each type of user in the user pool;
对于该用户池中每类用户下的某一个用户,将对应设备池中该用户未购买的设备生成推荐清单推荐给该用户。For a certain user under each category of users in the user pool, a recommendation list of devices that the user has not purchased in the corresponding device pool is generated and recommended to the user.
另外,本发明还提出一种智能家居设备的营销推荐系统,包括移动终端以及控制终端,移动终端包括输入输出设备和用于与控制终端通信的无线通信模块;控制终端包括用于与移动终端通信的无线通讯装置、存储器、处理器以及存储在所述存储器中并可在处理器上运行的计算机程序,所述处理器在执行所述计算机程序时实现上述的智能家居设备的营销推荐方法。In addition, the present invention also proposes a marketing recommendation system for smart home devices, including a mobile terminal and a control terminal, the mobile terminal including input and output devices and a wireless communication module for communicating with the control terminal; the control terminal including a wireless communication device for communicating with the mobile terminal, a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor implementing the above-mentioned marketing recommendation method for smart home devices when executing the computer program.
另外,本发明还提出一种智能终端,包括输入输出设备,存储器,处理器以及存储在所述存储器中并可在处理器上运行的计算机程序,所述处理器在执行所述计算机程序时实现上述的智能家居设备的营销推荐方法。In addition, the present invention also proposes an intelligent terminal, including an input and output device, a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned marketing recommendation method for smart home devices when executing the computer program.
有益效果是:本发明首先通过聚类算法对主机进行分类,得到每类主机的用户池,保证每个用户池具有相似的设备,接着采用聚类算法对用户池中的用户进行分类,在对用户池中的用户进行分类前,考虑到用户近期行为的影响,通过用户近期的行为采用时间衰减算法得到每个用户的推荐向量,进而根据推荐向量,采用聚类算法进行用户分类,使得每类用户中的用户更加相似,然后对用户池中的每类用户中的用户进行推荐时,考虑了用户拥有的设备。本发明通过双层聚类算法将推荐时,使用的数据范围进行缩小,同时兼顾用户近期的行为,不仅提高了推荐的准确性,而且有效的减少计算量,提高了推荐的效率,使得推荐的及时性提高。The beneficial effect is as follows: the present invention first classifies the hosts by a clustering algorithm to obtain a user pool for each type of host, ensuring that each user pool has similar devices, and then classifies the users in the user pool by a clustering algorithm. Before classifying the users in the user pool, the influence of the user's recent behavior is taken into account, and the recommendation vector of each user is obtained by a time decay algorithm based on the user's recent behavior. Then, based on the recommendation vector, a clustering algorithm is used to classify the users, making the users in each type of user more similar, and then when recommending the users in each type of user in the user pool, the devices owned by the users are taken into account. The present invention reduces the data range used when making recommendations by a double-layer clustering algorithm, while taking into account the user's recent behavior, which not only improves the accuracy of the recommendations, but also effectively reduces the amount of calculation, improves the efficiency of the recommendations, and improves the timeliness of the recommendations.
进一步的,上述智能终端、智能家居设备的营销推荐方法及系统中,为了提高时间衰减算法的准确性,不仅考虑了单日点击频次的影响力度,而且进一步考虑了同一设备的累积点击天数,所述时间衰减算法为:Furthermore, in the marketing recommendation method and system for the above-mentioned smart terminal and smart home device, in order to improve the accuracy of the time decay algorithm, not only the influence of the click frequency of a single day is considered, but also the cumulative number of click days of the same device is further considered. The time decay algorithm is:
其中,Y为用户的推荐向量,G为时间衰减系数,x1,x2,...,xT为距推荐日1天、2天、……、T天,用户每日的各设备点击次数向量,X为用户在T日内各设备的累计点击天数向量,α为调控X比重的权重系数。Where Y is the user's recommendation vector, G is the time attenuation coefficient, x 1 , x 2 , ..., x T are the vectors of the number of clicks on each device of the user every day 1 day, 2 days, ..., T days before the recommendation day, X is the vector of the cumulative number of clicks on each device of the user within T days, and α is the weight coefficient for adjusting the proportion of X.
进一步的,上述智能终端、智能家居设备的营销推荐方法及系统中,为了提高推荐的准确性,所述推荐清单为:设备池中点击频次高但用户未购买的设备所生成的推荐清单。Furthermore, in the marketing recommendation method and system for the above-mentioned smart terminal and smart home device, in order to improve the accuracy of the recommendation, the recommendation list is: a recommendation list generated by devices in the device pool that have a high click frequency but are not purchased by the user.
进一步的,上述智能终端、智能家居设备的营销推荐方法及系统中,为了减小计算量,统计每类主机下的用户时,过滤掉主机解绑的用户。Furthermore, in the marketing recommendation method and system for the above-mentioned smart terminal and smart home device, in order to reduce the amount of calculation, when counting the users under each type of host, the users who have unbound the host are filtered out.
进一步的,上述智能终端、智能家居设备的营销推荐方法及系统中,为了提高对主机分类的准确性,采用CURE聚类算法对主机进行分类。Furthermore, in the marketing recommendation method and system for the above-mentioned smart terminal and smart home device, in order to improve the accuracy of host classification, the CURE clustering algorithm is used to classify the hosts.
进一步的,上述智能终端、智能家居设备的营销推荐方法及系统中,为了提高用户分类的准确性,采用DBSCAN聚类算法对用户池中的用户进行分类。Furthermore, in the marketing recommendation method and system for the above-mentioned smart terminal and smart home device, in order to improve the accuracy of user classification, the DBSCAN clustering algorithm is used to classify the users in the user pool.
附图说明BRIEF DESCRIPTION OF THE DRAWINGS
图1为本发明智能家居设备的营销推荐方法的流程图。FIG1 is a flow chart of a marketing recommendation method for smart home devices according to the present invention.
具体实施方式DETAILED DESCRIPTION
智能家居设备的营销推荐系统实施例:Example of marketing recommendation system for smart home devices:
本实施例提出的智能家居设备的营销推荐系统,包括移动终端以及控制终端,移动终端包括输入输出设备和用于与控制终端通信的通信模块;控制终端包括用于与移动终端通信的通讯装置、存储器和处理器,存储器中存储有可在处理器上运行的计算机程序,处理器在执行计算机程序时实现智能家居设备的营销推荐方法。The marketing recommendation system for smart home devices proposed in this embodiment includes a mobile terminal and a control terminal. The mobile terminal includes an input and output device and a communication module for communicating with the control terminal; the control terminal includes a communication device, a memory and a processor for communicating with the mobile terminal. The memory stores a computer program that can be run on the processor. The processor implements the marketing recommendation method for smart home devices when executing the computer program.
这里的移动终端是一种为用户提供操作界面的终端,可以是android和ios等形式的终端设备,其中包括有基于手机、pad等电子产品的智能家居App,输入输出设备是触屏,控制终端为智能家居主机(以下简称主机),所实现的智能家居设备的营销推荐方法,将所得到的推荐列表发送至移动终端进行显示。当然控制终端也可以为主机加后台(为本实施例中的控制终端),实现控制终端相应的功能即可。而且关于通信模块,这里可以是蓝牙模块、WIFI模块、或者3G、4G、GPRS等,甚至也可以为有线通信模块,本发明对通信模块的具体实现方式不做限制。The mobile terminal here is a terminal that provides an operation interface for the user, which can be a terminal device in the form of Android and iOS, including smart home apps based on electronic products such as mobile phones and pads, the input and output devices are touch screens, and the control terminal is a smart home host (hereinafter referred to as the host). The marketing recommendation method of the smart home device implemented sends the obtained recommendation list to the mobile terminal for display. Of course, the control terminal can also be a host plus a background (the control terminal in this embodiment) to realize the corresponding functions of the control terminal. And about the communication module, it can be a Bluetooth module, a WIFI module, or 3G, 4G, GPRS, etc., or even a wired communication module. The present invention does not limit the specific implementation method of the communication module.
智能家居设备的营销推荐方法,包括如下步骤:The marketing recommendation method of smart home devices includes the following steps:
根据每台主机所绑定的设备清单,采用聚类算法对所有主机进行分类;Based on the device list bound to each host, a clustering algorithm is used to classify all hosts;
统计每类主机下的用户,构成每类主机的用户池;Count the users of each type of host and form a user pool for each type of host;
对于某一类主机的用户池,根据用户近期的设备点击次数,采用时间衰减算法得到该用户池中每个用户的推荐向量;根据该用户池中每个用户的推荐向量,采用聚类算法对该用户池中的用户进行分类;For a user pool of a certain type of host, a time decay algorithm is used to obtain a recommendation vector for each user in the user pool based on the number of recent device clicks by the user; based on the recommendation vector of each user in the user pool, a clustering algorithm is used to classify the users in the user pool;
统计该用户池中每类用户所在主机绑定的设备清单,构成该用户池中每类用户的设备池;Count the device list bound to the host of each type of user in the user pool to form a device pool for each type of user in the user pool;
对于该用户池中每类用户下的某一个用户,将对应设备池中该用户未购买的设备生成推荐清单推荐给该用户。For a certain user under each category of users in the user pool, a recommendation list of devices that the user has not purchased in the corresponding device pool is generated and recommended to the user.
具体的,智能家居设备的营销推荐方法如图1所示,包括如下步骤:Specifically, the marketing recommendation method for smart home devices is shown in FIG1 , and includes the following steps:
1)首先收集用户数据,并且统计所有出售的设备的细分种类(有a种),对设备种类进行编号。1) First, collect user data and count the subdivision categories of all devices sold (there are a types), and number the device types.
需要收集的数据有:每个主机下绑定的设备清单;智能家居所有设备清单及划分的细类;用户点击设备的操作记录;用户与主机绑定和解绑的记录。收集数据是通过主机定时上报后台(即云端),推荐方法的计算是在后台进行的,并将推荐结果保存至后台,在需要进行推荐时,后台将推荐结果下发至相应的主机,进而实现每个主机的推荐。The data that needs to be collected include: a list of devices bound to each host; a list of all smart home devices and their subcategories; user operation records of clicking on devices; and records of user binding and unbinding with hosts. The collected data is reported to the background (i.e., the cloud) by the host at regular intervals. The calculation of the recommendation method is performed in the background, and the recommendation results are saved in the background. When a recommendation is needed, the background will send the recommendation results to the corresponding host, thereby realizing the recommendation of each host.
2)根据每台主机所绑定的设备清单,生成每个主机的特征向量Q,特征向量Q中每个分量分别代表a种设备的购买数量,若未购买某种设备,则该分量值为0,对所有主机生成的特征向量采用聚类算法进行分类,形成n类主机。2) Generate a feature vector Q for each host based on the list of devices bound to each host. Each component in the feature vector Q represents the number of a type of equipment purchased. If a certain device is not purchased, the component value is 0. The feature vectors generated by all hosts are classified using a clustering algorithm to form n types of hosts.
聚类算法就是按照某个特定标准(如距离准则)把一个数据集分割成不同的类或簇,使得同个簇内的对象的相似性尽可能大,不在同簇中的对象的差异性尽可能大。A clustering algorithm divides a data set into different classes or clusters according to a specific standard (such as distance criterion), so that the similarity of objects in the same cluster is as large as possible, and the difference of objects in different clusters is as large as possible.
本实施例中,对主机分类采用的聚类算法为CURE聚类算法,属于层次聚类法中的凝聚法,可解释性好,该聚类算法能产生高质量的聚类,抗异常值能力强,能解决非球形族。该聚类算法初始迭代前需提供4个输入数据:数据集D,分类数目K,收缩因子a取0.2-0.7,代表点数c一般取10以上效果佳。作为其他实施方式,对主机的分类也可以采用其他聚类算法,例如WAVECLUSTER、ROCK、BIRCH、K-PROTOTYPES、DENCLUE、OPTIGRID、CLIQUE、DBSCAN、CLARANS等,本发明对此不做限制。In this embodiment, the clustering algorithm used for host classification is the CURE clustering algorithm, which belongs to the agglomerative method in the hierarchical clustering method and has good interpretability. The clustering algorithm can produce high-quality clusters, has strong anti-outlier capabilities, and can solve non-spherical families. Before the initial iteration of the clustering algorithm, four input data need to be provided: data set D, number of classifications K, shrinkage factor a is 0.2-0.7, and the number of representative points c is generally 10 or more for best results. As other implementation methods, other clustering algorithms can also be used for the classification of hosts, such as WAVECLUSTER, ROCK, BIRCH, K-PROTOTYPES, DENCLUE, OPTIGRID, CLIQUE, DBSCAN, CLARANS, etc., and the present invention does not limit this.
3)统计n类主机中每类主机下的用户,构成每类主机的用户池,总计n个用户池。3) Count the users of each type of host among the n types of hosts to form a user pool for each type of host, totaling n user pools.
主机下的用户包括与主机绑定过的用户,因此在该步骤中应当对用户进行过滤,判断用户与主机是否解绑,对于解绑的用户或者绑定时间小于T日的用户,不再进行统计推荐,减少计算量;对于一直绑定或者T日内重新绑定的用户,进行统计推荐。而且T日内重新绑定的用户,其数据从绑定之日起记录统计。Users under a host include those who have been bound to the host. Therefore, users should be filtered in this step to determine whether the user is unbound from the host. For unbound users or users whose binding time is less than T days, no statistical recommendation will be made to reduce the amount of calculation; for users who have been bound or rebound within T days, statistical recommendation will be made. Moreover, for users who are rebound within T days, their data will be recorded and counted from the date of binding.
4)对于某一类主机的用户池,根据用户近期T日的设备点击次数(即操作记录),采用时间衰减算法得到该用户池中每个用户的推荐向量;然后根据该用户池中每个用户的推荐向量,采用聚类算法对该用户池中的用户进行分类。分为m类。4) For a user pool of a certain type of host, according to the number of device clicks (i.e., operation records) of the user in the recent T days, the time decay algorithm is used to obtain the recommendation vector of each user in the user pool; then, according to the recommendation vector of each user in the user pool, the clustering algorithm is used to classify the users in the user pool into m categories.
本实施例中,为了提高时间衰减算法的可靠性,时间衰减算法为:In this embodiment, in order to improve the reliability of the time decay algorithm, the time decay algorithm is:
其中,Y为用户的推荐向量(也即推荐日的预测向量),G为时间衰减系数,x1,x2,...,xT为距推荐日1天、2天、……、T天,用户每日的各设备点击次数向量,X为用户在T日内各设备的累计点击天数向量,α为调控X比重的权重系数。Where Y is the user's recommendation vector (i.e., the prediction vector of the recommendation day), G is the time attenuation coefficient, x 1 , x 2 , ..., x T are the vectors of the number of clicks on each device of the user every day 1 day, 2 days, ..., T days before the recommendation day, X is the vector of the cumulative number of clicks on each device of the user within T days, and α is the weight coefficient for adjusting the proportion of X.
作为其他实施方式,也可以采用现有技术中的时间衰减算法得到推荐清单,现有时间衰减算法为:As another implementation, the recommendation list may also be obtained by using a time decay algorithm in the prior art. The prior time decay algorithm is:
该公式中的字符解释与上述公式相同,这里不做赘述。The interpretation of the characters in this formula is the same as that in the above formula and will not be repeated here.
现有的时间衰减算法侧重点是对单日点击频次的影响力度的考量,然而根据实际用户操作情况,同一设备的累积点击天数也尤为重要,因此本发明提出的时间衰减算法考虑了同一设备的累积点击天数,将现有技术进行优化,而且参数G和α的确定主要根据测试数据、精确率和召回率进行确定,确定过程为现有技术,这里不做过多赘述。The existing time decay algorithm focuses on the consideration of the impact of a single day's click frequency. However, according to actual user operations, the cumulative number of days of clicks on the same device is also particularly important. Therefore, the time decay algorithm proposed in the present invention takes into account the cumulative number of days of clicks on the same device, optimizes the existing technology, and the parameters G and α are mainly determined based on test data, precision and recall. The determination process is existing technology and will not be elaborated here.
本实施例中,得到每个用户的推荐向量后,采用基于密度的聚类算法——DBSCAN聚类算法对该用户池中的用户进行分类,该算法聚类速度快且能够有效处理噪声点和发现任意形状的空间聚类,并且不需要指定分类数目,算法自行确定。作为其他实施方式,也可以采用其他聚类算法进行用户的分类,例如WAVECLUSTER、ROCK、BIRCH、K-PROTOTYPES、DENCLUE、OPTIGRID、CLIQUE、CURE、CLARANS等。In this embodiment, after obtaining the recommendation vector of each user, the density-based clustering algorithm, DBSCAN clustering algorithm, is used to classify the users in the user pool. The algorithm has a fast clustering speed and can effectively process noise points and find spatial clusters of arbitrary shapes. It does not need to specify the number of classifications, and the algorithm determines it by itself. As other implementation methods, other clustering algorithms can also be used to classify users, such as WAVECLUSTER, ROCK, BIRCH, K-PROTOTYPES, DENCLUE, OPTIGRID, CLIQUE, CURE, CLARANS, etc.
聚类算法用到的样本点之间的度量公式主要有曼哈顿距离、欧式距离、闵科夫斯基距离,马氏距离等,若未特别指定,一般均默认采用欧几里得度量,欧几里得度量(euclidean metric)也称欧氏距离。在数学中,欧几里得距离或欧几里得度量是欧几里得空间中两点间“普通”(即直线)距离,在欧几里得空间中,对于n维的两个样本点x=(x1,x2,...,xn)和y=(y1,y2,...,yn)的欧式距离Dist(x,y)为:The metric formulas between sample points used in clustering algorithms mainly include Manhattan distance, Euclidean distance, Minkowski distance, Mahalanobis distance, etc. If not specified, Euclidean metric is generally used by default. Euclidean metric is also called Euclidean distance. In mathematics, Euclidean distance or Euclidean metric is the "ordinary" (i.e. straight line) distance between two points in Euclidean space. In Euclidean space, the Euclidean distance Dist(x,y) of two n-dimensional sample points x = (x 1 ,x 2 ,...,x n ) and y = (y 1 ,y 2 ,...,y n ) is:
5)统计该用户池中每类用户所在主机绑定的设备清单,构成该用户池中每类用户的设备池,对于该用户池中每类用户下的某一个用户,将对应设备池中该用户未购买的设备生成推荐清单推荐给该用户。5) Count the device lists bound to the hosts of each type of user in the user pool to form a device pool for each type of user in the user pool. For a certain user under each type of user in the user pool, generate a recommendation list of devices that the user has not purchased in the corresponding device pool and recommend them to the user.
本实施例中,每个设备池中,根据该类用户对设备的点击次数,对设备池中的设备进行降序排列,推荐给用户的推荐清单为:设备池中点击频次高但用户未购买的N个设备所生成的推荐清单。作为其他实施方式,也可以将该类用户中对设备的购买数量作为排序的标准,进而将购买数量对但用户未购买的设备进行推荐,本发明对此不做限制。In this embodiment, in each device pool, the devices in the device pool are sorted in descending order according to the number of clicks on the devices by this type of user, and the recommended list recommended to the user is: a recommended list generated by N devices in the device pool that have high click frequencies but are not purchased by the user. As another implementation, the number of purchases of devices by this type of user can also be used as a sorting standard, and then the devices with the purchase number but not purchased by the user can be recommended, and the present invention does not limit this.
本发明采用双层聚类算法,要明显优于单层聚类。有效节省了计算开销,因为单层聚类将同时面对主机数据和用户的操作记录,若将主机和用户数据构成矩阵,通过计算矩阵间的相似度来进行聚类,则显然高于双层聚类的分两次分别计算相量间的相似度的计算量,而且为了使得分类更加精细化,用户的分类不仅基于主机下的设备,也兼顾了用户的近期设备的点击操作习惯,让同类用户的相似度更高,从而实现有效的推荐。The present invention adopts a double-layer clustering algorithm, which is significantly better than single-layer clustering. It effectively saves computational overhead, because single-layer clustering will face both host data and user operation records at the same time. If the host and user data are formed into a matrix, clustering is performed by calculating the similarity between matrices, which is obviously higher than the computational cost of double-layer clustering, which calculates the similarity between phase quantities twice. In addition, in order to make the classification more refined, the classification of users is not only based on the devices under the host, but also takes into account the user's recent device click operation habits, so that the similarity of similar users is higher, thereby achieving effective recommendation.
智能终端实施例:Smart terminal implementation example:
本实施例提出的智能终端与智能家居设备的营销推荐系统的不同之处在于,智能家居设备的营销推荐系统中,控制终端对所采集的数据进行处理并生成推荐清单,将推荐清单在移动终端中进行显示,是两个可以进行通信、独立的设备,而智能终端是将采集、处理与显示集为一体的设备。The difference between the marketing recommendation system of the smart terminal and the smart home device proposed in this embodiment is that in the marketing recommendation system of the smart home device, the control terminal processes the collected data and generates a recommendation list, and displays the recommendation list in the mobile terminal. They are two independent devices that can communicate, while the smart terminal is a device that integrates collection, processing and display.
智能终端包括输入输出设备,存储器,处理器以及存储在存储器中并可在处理器上运行的计算机程序,处理器在执行所述计算机程序时实现智能家居设备的营销推荐方法。The intelligent terminal includes an input and output device, a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, a marketing recommendation method for the intelligent home device is implemented.
智能家居设备的营销推荐方法的具体实施过程在上述智能家居设备的营销推荐系统实施例中已经介绍,这里不做赘述。The specific implementation process of the marketing recommendation method for smart home devices has been introduced in the above-mentioned marketing recommendation system embodiment for smart home devices, and will not be repeated here.
智能家居设备的营销推荐方法实施例:Implementation example of marketing recommendation method for smart home devices:
本实施例提出的智能家居设备的营销推荐方法,包括以下步骤:The marketing recommendation method for smart home devices proposed in this embodiment includes the following steps:
根据每台主机所绑定的设备清单,采用聚类算法对所有主机进行分类;Based on the device list bound to each host, a clustering algorithm is used to classify all hosts;
统计每类主机下的用户,构成每类主机的用户池;Count the users of each type of host and form a user pool for each type of host;
对于某一类主机的用户池,根据用户近期的设备点击次数,采用时间衰减算法得到该用户池中每个用户的推荐向量;根据该用户池中每个用户的推荐向量,采用聚类算法对该用户池中的用户进行分类;For a user pool of a certain type of host, a time decay algorithm is used to obtain a recommendation vector for each user in the user pool based on the number of recent device clicks by the user; based on the recommendation vector of each user in the user pool, a clustering algorithm is used to classify the users in the user pool;
统计该用户池中每类用户所在主机绑定的设备清单,构成该用户池中每类用户的设备池;Count the device list bound to the host of each type of user in the user pool to form a device pool for each type of user in the user pool;
对于该用户池中每类用户下的某一个用户,将对应设备池中该用户未购买的设备生成推荐清单推荐给该用户。For a certain user under each category of users in the user pool, a recommendation list of devices that the user has not purchased in the corresponding device pool is generated and recommended to the user.
智能家居设备的营销推荐方法的具体实施过程在上述智能家居设备的营销推荐系统实施例中已经介绍,这里不做赘述。The specific implementation process of the marketing recommendation method for smart home devices has been introduced in the above-mentioned marketing recommendation system embodiment for smart home devices, and will not be repeated here.
Claims (7)
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010067606.2A CN112052380B (en) | 2020-01-20 | 2020-01-20 | Marketing recommendation method and system for smart terminals and smart home devices |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010067606.2A CN112052380B (en) | 2020-01-20 | 2020-01-20 | Marketing recommendation method and system for smart terminals and smart home devices |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112052380A CN112052380A (en) | 2020-12-08 |
CN112052380B true CN112052380B (en) | 2024-08-27 |
Family
ID=73609145
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010067606.2A Active CN112052380B (en) | 2020-01-20 | 2020-01-20 | Marketing recommendation method and system for smart terminals and smart home devices |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112052380B (en) |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108052639A (en) * | 2017-12-21 | 2018-05-18 | 中国联合网络通信集团有限公司 | Industry user based on carrier data recommends method and device |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8095521B2 (en) * | 2007-03-30 | 2012-01-10 | Amazon Technologies, Inc. | Recommendation system with cluster-based filtering of recommendations |
CN105426548A (en) * | 2015-12-29 | 2016-03-23 | 海信集团有限公司 | Video recommendation method and device based on multiple users |
CN109426977A (en) * | 2017-08-28 | 2019-03-05 | 北京嘀嘀无限科技发展有限公司 | A kind of information processing method, information processing system and computer installation |
CN110517114A (en) * | 2019-08-21 | 2019-11-29 | 广州云徙科技有限公司 | A kind of information-pushing method and system based on community discovery algorithm |
-
2020
- 2020-01-20 CN CN202010067606.2A patent/CN112052380B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108052639A (en) * | 2017-12-21 | 2018-05-18 | 中国联合网络通信集团有限公司 | Industry user based on carrier data recommends method and device |
Also Published As
Publication number | Publication date |
---|---|
CN112052380A (en) | 2020-12-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105320957B (en) | Classifier training method and device | |
CN113468227B (en) | Information recommendation method, system, equipment and storage medium based on graph neural network | |
CN105224699B (en) | News recommendation method and device | |
CN111143697A (en) | Content recommendation method and related device | |
CN109345368A (en) | Credit estimation method, device, electronic equipment and storage medium based on big data | |
CN105005589A (en) | Text classification method and text classification device | |
CN109242002A (en) | High dimensional data classification method, device and terminal device | |
CN110895706B (en) | Method and device for acquiring target cluster number and computer system | |
CN107480696A (en) | A kind of disaggregated model construction method, device and terminal device | |
CN118503889B (en) | Financial risk prevention and control large model data processing method and electronic equipment | |
WO2022111095A1 (en) | Product recommendation method and apparatus, computer storage medium, and system | |
CN113159213A (en) | Service distribution method, device and equipment | |
Ma et al. | A spectral method for assessing and combining multiple data visualizations | |
CN113169890A (en) | Fast and efficient classification system | |
CN116539994A (en) | Substation main equipment operation state detection method based on multi-source time sequence data | |
CN112052380B (en) | Marketing recommendation method and system for smart terminals and smart home devices | |
CN114461899A (en) | Collaborative filtering recommendation method and device for user, electronic equipment and storage medium | |
CN113988149A (en) | A Service Clustering Method Based on Particle Swarm Fuzzy Clustering | |
CN108171538A (en) | User data processing method and system | |
CN108647263B (en) | Network address confidence evaluation method based on webpage segmentation crawling | |
CN110837853A (en) | A Fast Classification Model Construction Method | |
CN118733871A (en) | Information recommendation method, device, electronic device and storage medium | |
CN114547482A (en) | Service feature generation method and device, electronic equipment and storage medium | |
Zeng et al. | An Enhanced Gas Sensor Data Classification Method Using Principal Component Analysis and Synthetic Minority Over-Sampling Technique Algorithms | |
CN114511016A (en) | Model-based adaptive adjustment user attribution prediction method, device and equipment |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |