As we know that user preferences get change with the passage of time, in that case, for how long time a user profile should be updated in order to achieve better recommendation results and user satisfaction. Also, if a user preferences changes rapidly then, for how long system should keep the previous information/history of the user and consider these information in recommendation process.
Dear Zafar,
User profile should be update forever. We can use currently update profile to predict user preferences, an ad hoc or transient user needs. Profile history can be used to develop user background knowledge.
User background knowledge is used to predict user need. For example, a Java programmer has a background knowledge in Java programming language. We can use it in the recommendation process if there is no current update on user profile. But, whenever the user looking for hotels in Java island, then the recommender system use currently update of user's profile.regards,
Harry
Conference Paper Web Mining Based on User Profile and Preferences
Brother, Update users' profile(s) as soon as new information about such users are available. This gives your system high precision. However if you consider this might affect other parts of your system make it periodic with small intervals
There are users that frequently interact with the system while others are less interactive and visit your system less frequently. Therefore, if you keep short time intervals, the less frequent users will suffer. Also, how can you categorize user history between previous and current, I mean what is the criteria that makes this division. Therefore, the profile should be updated as soon as new data about the user is available.
In case of content-based recommenders, you may study this paper:
Improving Rocchio Algorithm for Updating User Profile in Recommender Systems byChong Wang, Yao Shen, Huan Yang, Minyi Guo
While in case of collaborative or hybrid filtering, you can find several relevant papers using Google Scholar.
Hope this helps...
Irfan Ullah,
I appreciate your answer, but when a user interact with a system very frequently and her preferences changes abruptly then what should we do in that case? weather we should keep track of her previous preferences or we should only focus on the current information and needs? if we want to keep previous information in consideration then what could be the time threshold for that?
I get your questions Zafar... You can scale the interest rate with a time decay function(negative exponential of time). This is to make users older interests have lesser contribution in the recommendations than the more current interests. This is time-dependent recommendation. check my paper:
Article Modelling a Collaborative Digital Research Ecosystem Based o...
- Badges
- Science topic