We will find a lot of sites with content recommended, not only is recommended excellent books like B2C e-commerce class, but also interested in sites guess like watercress Douban. Such functions undoubtedly played a significant effect to help users find the demand, promotion of commodity purchase and service applications. Such recommendation is how to get it? In fact, with the site data analysis, we can simply look at its principles and implementation.
The association recommended marketing is divided into two categories:
Up Marketing (Up Marketing): according to consumer preferences for existing customers, provide higher value to strengthen its original function or use products or services.
Cross-marketing (Cross Marketing): the diverse needs of customer buying behavior of customers to sell related products or services.
Up marketing is based on a similar product line, upgrade or optimize the product recommended, and cross-marketing is based on the recommendation of the similar but different types of products. Give a simple example, look at Apple's product line:
When you buy an ipod nano3 to recommend to you to upgrade the product nano4, nano5 or function similar to the itouch called up marketing; recommended Iphone, Mac, or ipad is the "cross-marketing".
The association recommended that the implementation can be divided into two types: the associated recommendation based on product analysis and user analysis recommended based association. Product analysis related recommendation refers to are the Web site by analyzing the characteristics of the product found in common between them, such as Web Analytics and the Web Analytics 2.0 "the author Avinash Kaushik, and the title contains the Web Analytics, analysis class books, but also a Press ... then based on the relevance of the product can be recommended to users who purchased Web Analytics the Web Analytics 2.0 ". Analysis based on user recommendations by analyzing the user's historical behavior data may be found to purchase the many users of the Web Analytics also bought the book "The Elements of User Experience, then recommended based on this discovery, this method is mining association rules in data mining (Association, the Rules), one of the most classic example is Wal-Mart of beer and diapers story.
Many of the association recommended or products based on the product level, because the implementation is much simpler (for the site in terms of product data significantly less than the user behavior data, and may differ by several orders of magnitude, so the analysis will be much lighter), based on the recommendation of more above-described two kinds of marketing tools to achieve more biased in favor of the traditional "push" marketing (personal do not like this type of marketing, especially the "bundling" and the like).
Associated recommendation based on user behavior analysis
Personally biased in favor of the analysis based on user, which is more conducive to the discovery of the potential demand of the user, to help users better choice of products they need, by the user decide whether to buy, that is, the so-called "pull" marketing. Products or services by referring users to stimulate the potential needs of the user, prompting the user consumption more in line with the "user-centric" concept. Following a brief description of user behavior analysis based association recommend, whether you are e-commerce site or any other type of site, in fact, this feature can be achieved, as long as you have the following premise:
- Able to efficiently identify the user;
- Retains the user's historical behavior data (click stream data (clickstream) or operational data (outcomes));
- Of course, also need a good site data analyst.
Here to e-commerce site as an example to illustrate the concrete realization of what association rules. Currently, most e-commerce site to provide the functionality of the user registration, shopping, users are generally the conditions based on login, so here are provided for the user to identify the most effective identifier - the user ID (user identification methods Please refer to the article - the identification of the user ); website will all the user's shopping data stored in its own operations inside the database, this data base - User historical shopping data for the analysis of user behavior. So meet the first two conditions above, we can proceed with the analysis.
The principle of association rules from the data of all users shopping (if the data is too large, you can select a certain time interval, such as year, quarter, etc.) to find when the user has purchased the goods A bought B proportion of the number of commodities, when this ratio reached a target level of default, we believe that these two commodities there is a certain correlation, so when the user has purchased product A but not yet purchased the B Product B Product that we can recommend to such users. As shown below:
Involved three sets: all users Complete Works of U-purchased goods, purchased the goods A user sets A and B Product A commodity purchase buy set G can be seen from the figure. Based on two key indicators in the three collections can be calculated association rule mining - support (Support) and confidence (the Confidence):
The number of support = the number of people to buy the goods of A and B (set G) / all purchased goods (set U)
Confidence = the number of people to buy the goods of A and B (set G) / purchase of goods (A collection of A) the number of
These two indicators, the need to establish a minimum threshold for these two indicators, the minimum support and minimum confidence. Also purchased in customer acquisition, the purchase of goods A user may not only buy B goods, C, D, E, and so on a range of commodities, so we need to calculate the support and confidence of all these combinations, only meet such as support for the> 0.2 confidence level> 0.6 for these commodities portfolio can think there is a link, it is recommended.
Of course, if your site is not e-commerce site, you can also use a user browses the site click-stream data associated with the recommended features. The same behavior based on user history, such as browse page A user can also browse the B page, watch the video A user watched the B video, download the file A user downloaded the B file ...
Mining association rules in data mining, generally use the Apriori algorithm based on frequent itemsets, which is a relatively simple and effective algorithm, where not specifically introduced, and friends who are interested can go to check the following information.
During the association rules need to pay attention to some of the problems
- Note the association recommended the scope and prerequisites, not every type of site are appropriate or need to be associated with the recommended;
- Established in the minimum support and minimum execution need to be set according to the characteristics of Web site operators should not be high or low, recommendations based on the experiment or practice on the basis of continuous optimization to find an optimal trade-off point.
- Require special attention, in the association rule A commodity and commodity B, does not mean that the B goods and commodities A correlation was established because of the confidence of both algorithms are different, the associated direction irreversible.
- Association rules in the algorithm is not difficult, but to be on the site truly good, meet the three above premise on the basis of the need for continuous optimization algorithm, and need the collaboration of various departments of the site to achieve .
Therefore, the association recommended based on user behavior analysis completely from the user's point of view to analyze further and more effective than the simple comparison between products associated more closely match user behavior, to discover the potential demand of the user may wish to try.
»In this paper, the BY-NC-SA agreement, reproduced, please indicate the source: The data analysis » up marketing, cross marketing and associated recommended
Related Articles:
In fact, a rigid division starting from the product "and" starting from the user "of little significance, in essence, the correlation between the analysis of product (or similar), just on the relevance of the definition of different emphases. These two strategies should be unified. In addition, to enhance the degree of association rules is worth it to consider one of the indicators.
@ Phi Phi Miffy : is not rigidly divided into "starting from the product" and "departure" from the user, but would like to express more from a user perspective to understand the user's behavior and habits, and so are better able to meet the needs of users.
Enhance the degree is indeed a good measure.
The original bloggers the zju alumni mentors, admirers admire ~ ~ ~
Here whether there is a clerical error, "the purchase of all goods (set U), the number should be the number of users purchased goods?
_AT_ Franc : Well, the phrase is indeed a problem, I changed, I would like to thank the remind!
What is your drawing tool is? ? ? Character has?
@ Aibei Fu : This chart is of my own PS, technology is not good, laughed
A lot of things learned from the blogger's blog, has been diving. Yesterday decided to first read the study, see the article to know the original blogger is zju graduated, so I decided to leave footprints, to express my admiration of this ZJUer seniors ~ ~
Leona : Oh, glad to here can encounter many of our alumni.
Apple this end-to-end success stories are difficult to reproduce Na. . . . .