First, the need for user behavior analysis
User behavior analysis is the basis of many algorithms designed early as statistical popular sort, although simple but very popular with most users prefer, because it saves the user can easily find their favorite things time. Later, the algorithm is recommended as a more in-depth analysis of user behavior, it can bring a better experience to the user.
Second, the kind of user analysis
User behavior analysis according to different angles can be divided into four categories, according to which explicit feedback may be divided into explicit feedback and implicit feedback, and can be divided according to the direction of positive feedback and negative feedback.
| Explicit feedback | Implicit Feedback | |
| Positive feedback | A | C |
| Negative feedback | B | D |
Wherein the explicit feedback indicates that the user define their preferences, such as a movie score assessment, implicit feedback indicates that the user does not define their preferences, such as the purchase of sports magazines implicitly expressed user behavior like sports. Positive feedback is the user like behavior, negative feedback is that users do not like the behavior.
On the table
A: articulate about something they like behavior
B: articulate about something they do not like the behavior
C: recessive express themselves about something like behavior;
D: recessive expression about something they do not like the behavior;
Third, how to describe user behavior
1, in most cases the Internet with a standard to measure user behavior is difficult, but you can use the following dimensions to describe the behavior of a user normally
| user_id | Behavior of user-generated unique identification |
| item_id | Uniquely identifies the behavior of the object to produce |
| behavior_type | The behavior of species such as the purchase, browse, click |
| context | Generating behavior context, such as time, place |
| behavior_weight | I.e., the weight acts describe the behavior characteristic values as quantization look length |
| behavior_context | Behavioral content, such as comments, scoring |
Of course field behavior is described in the table according to their needs change, a few small fields or add a few fields are possible, there is no standard answer only the best for their own analysis.
2, in the Internet field, after a lot of research studies have found that user behavior data there is a general rule, these laws meet the long-tailed distribution
What are the long-tailed distribution, long-tailed distributions My understanding is that the more things appear less frequency of its kind, as most of the English word word frequency is actually very low, frequently used words are very few.
In fact, there are long-tailed distribution of user behavior data, which can be user activity and the popularity of items to represent.
| User Activity | The total number of user-generated articles over behavior |
| Popularity items | The number of users of goods produced a behavior |
Researchers describe below shows the distribution of popular articles, wherein the article is an article popularity abscissa, ordinate the total number of items corresponding to the coordinates of popularity, it can be seen from the figure, the less the total number the higher the popularity of the article, this is a good description of a small number of popular shopping sites in total merchandise goods, most goods are not getting attention.
The figure is described with a distribution of user activity, user activity in which the abscissa, the ordinate is the total number of people corresponding activity, activity can be obtained from the figure the better the higher the total number, which is well described the number of active members of the site group always, most of them are relatively small part of the silent member.
The relationship between user activity and welcome of the items can be seen by the figure below
As can be seen from the figure, the higher the degree of active users with its more have the opportunity to welcome those who come into contact with items not high for a low user activity such as new customers or clients that prefer to focus on the silence of welcome higher items, this study can give recommendations on how to better provide product ideas.
