Tasks

1. Fix minor issues in the document.
2. Start doing some simple experiments with the item selection to see if
• discussed methods are able to suggest items that the user knows
• acquisition of this information produces changes in the recommendation list.
3. Add kendale t to the list of methods for comparing lists.
4. Revise the strategy for improving the reliability of neighbors' similarity.
5. Add the strategy for eliciting opinion on items correlated to the best current recommendations.
6. Compute the p value using a different approach for ? (unknown) And ø (a user doesn't have and opinion).
7. Read the Carenini paper one more time and understand the simulation part completely.
8. Look at the abstract that my supervisor has sent for useful papers.
9. Look at the literature for predicting if a user has/doesn't have an opinion about an item.

Progress

• Some refactoring for the text
• Added/finished the following strategies: most controversial, entropy, items with worst/best predictions

TODO:

• read it one more time in the morning.
• talk the professor and start testing.
• acquire the rating matrix for testing.
• ask again about the strategy of "items more correlated with best predictions". What is exactly correlation between two items? is it pearson/cosine similarity? something else?
  

Progress

• Wrote a chapter defining strategy for improving the neighborhood.
• Found a nice paper on using entropy and other methods for eliciting new users: Getting to know you: learning new user preferences in recommender systems

Todo: rename chapters. Now there's a real mess with strategy names repeating. Also it's not clear where we use the p * f and where we don't.

Progress and Notions

• Fixed minor issues in the document.

• Corrected the utility computation formula. I left it simply p*f.

What I felt I should/shouldn't do:

• Keep it simple
• Tests first, (formula, algorithm) optimizations later
• Start simple testing soon: get real. Otherwise I will be stuck in an endless loop of modifying and refining descriptions of the ideas that I'm not sure work at all.
  

Progress

• Converted the draft documents to MS Office format. Some problems with formulas and tables had to be solved. Also enabled the tracking of changes.
  

Progress

• Read the Carenini paper "Towards more Conversational and Collaborative Recommender Systems" again. There are things in the simulation part that I still don't understand, will have to look at it later again.
• Installed Microsoft Office 2007.

Current Tasks

1. Fix minor issues in the document.
2. Correct the utility computation formula.
3. List other possible ways to compute the effect of a new rating in the output.
4. Include some methods for item selection not based on the effect of the rating on the prediction:
   • entropy
   • increase overlap with users
   • most controversial
   • items more correlated with best predictions
   • items with worst predictions
   • items with best and with worst predictions
5. Compute the p value using a different approach for ? (unknown) And ø (a user doesn't have and opinion).
6. Read the Carenini paper more carefully.
7. Look at the abstract that my supervisor has sent for useful papers.
8. Look at the literature for predicting if a user has/doesn't have an opinion about an item.
9. Start doing some simple experiments with the item selection to see if
   • discussed methods are able to suggest items that the user knows
   • acquisition of this information produces changes in the recommendation list.