Implicit mining of Web site user behavior

Reference: http://ucdchina.com/blog/?p=12

How to understand the needs of users. According to the user whether the active participation is divided into explicit and implicit two types of mining mode, because the explicit movement is relatively big, has the very big limitation, therefore in order to guarantee the result accuracy as well as enhances the user acceptance degree, generally uses the implicit type.

The user's daily interaction behavior will produce four kinds of key data: The mouse moves the track, the link clicks the distribution, the page browsing flow, the page stays time. Through the user's behavior can reflect the user's point of view, while using the access to the page order can find the hidden relationship between the pages.

Collect Data Web server log (user session record) Web trends or similar Third-party shareware (client analysis, traffic analysis, usability analysis) Third party software/plug-ins developed by oneself (requirement customization)

Large Web sites will usually put the above three methods combined application, the general principle is to enter the site to give users identity recognition, each generated interactive action to the server to send back the request, through the time and page to determine the connection to each request point and recorded. (algorithm not discussed)

Filter Data clear targets, define core data. Define user behavior, use the behavior of most people to eliminate the subjectivity of individual behavior. Categorize the user and determine the category of data.

The amount of data generated by a large web site every day is staggering, so conventional requirements are usually timed or quantitatively analyzed. In addition, the additional data processing will slow down the speed of the Web site, the more information collected, the greater the potential negative impact.

Custom Analysis of the user visited the page for content analysis, according to the information theme of the page clustering. In addition to considering the similarity of the page content in the clustering process, the page path should also be considered. The effect of user browsing behavior on its interest is included in the clustering results, and the comprehensive evaluation model is obtained.

The interest of the user is divided into two situations, which are accidental and stable, which can be thought to be random change, stable excavation and based on the content and behavior of two ways, in the content of the performance of the repetition, similarity, etc., in the behavior of the performance of the length of stay, the number of points, pull the number of scroll bar.

actual Case

Similar system, browser, resolution client analysis, common and simple, skip.

About the mouse trajectory, click on the availability of distribution examples: tracking the user in the search of the mouse movement trajectory, you can obtain user action sequence, hotspot function, action curve and other primary data, these are the important reference to improve or simplify the form. In the important page for detailed click Distribution Monitoring statistics, mainly check the information presented by the ease of use, to see if there is no deviation from the original design, often updated to find the law.

To deal with specific user behavior, user groups, user antecedents of the task flow example: monitoring the distributed registration process, can see how many users fill out the form, fill out the form, or in a certain step of abnormal loss. Monitor the entry of different modules of registered users, to be able to calculate the number of modules imported effective registration, percentage, success rate, in order to reasonably allocate resources. Monitor the number of registered ads, registration success rate, conversion rate of pay users, in order to clear the input-output ratio of advertising. Monitoring user's depth browsing behavior is a good way to test navigation usability, which means that users will not get lost in your site.

End

Again, the discussion here is limited to the implicit mining of later research, that is, not to alarm users, not to allow users to detect the method, completely through technical means to get the data we want to know, and then through theoretical analysis to improve the product.

Prior to a listed company full participation in the development of similar systems, the hardest part of how to obtain data in the basic construction phase, but after all is a matter of once and for all, beneficial to the future development of harmless.