Hi,
I have a set of users that each user contains a vector of some parameters, and I want to develop a recommender system to recommend an item to each user. My goal is to change users behaviors, it means that the recommender systems should not be trained based on previous users preferences data.
My problem is that how I can generate a huge number of items for training the model because I have many users and I cannot check out all users person by person.
Do you know any solution for this problem?
Dear Ashkan Amirnia , it would be interesting to know what information is kept in the vector. Mainly, to think about the relationship among the available data and the kinds of recommendations that the system should provide. If you want to use a supervised model, you should provide at least an initial data set. After that, depending on the chosen model, you could incrementally go updating it. It could be a good idea to take a look at the Massive Online Analysis (MOA) project (http://moa.cms.waikato.ac.nz) and Stream Mining. It could be useful for orienting your perspective. Regards
Dear Mario José Diván,
Thank you for your interesting answer. Is it possible to create some items for training data by a rule-based program based on each vector's situation, then use them for training a supervised model?
Regards,
Ashkan
To do recommendations you need to use past experiences then only you can train model. If you want to incorporate user changing behavior, use time evolution factor. Meaning , keep user data in specific time slice and keep updating it.
following paper at link might help you,
https://www.mdpi.com/2078-2489/11/1/4/pdf
You can try parameter sweeping approaches that auto tune the parameters based on the uses' behaviors for performance optimization.
Conference Paper On Parameter Tuning in Search Based Software Engineering
Dear Ashkan Amirnia even when it is possible you need to be careful of it. It is possible to do it as a proof-of-concept, however, the generated data do not represent an actual characterization of some situation or concept. It is recommended to obtain some actual datasets representing a given situation based on the properties required by the recommendation system.
- Badges
- Science topic