My paper entitled “Personalization of Content Ranking in the Context of Local Search” has been accepted for publication at the IEEE/WIC/ACM International Conference on Web Intelligence (WI2009) as a regular paper. The acceptance rate was pretty low at 16%. The conference will be co-located with the Intelligent Agent Technology Conference (IAT2009) in Milan, Italy. My colleague and co-author Philip O’Brien and I will be presenting the paper. The other co-authors are Xiao Luo, Weizheng Gao, and Shujie Li. The conference will be held in September 15-18, 2009 at Università degli Studi di Milano Bicocca.
If you’re planning to attend WI/AIT 2009 and would like to connect, contact me.
The abstract of the paper follows:
Ranking search results using a single ranking function for all search engine visitors is inherently bounded in the performance the ranking algorithm can achieve when considering the variety of requirements of Web searchers and the proliferation of topics and types of data modern search engines rank. Adding a geographical dimension to the mix by way of local search engines further reduces the average satisfaction a ranking algorithm can garner from local search users. Personalization has been proposed in Web search with some success but has not, to our knowledge, been investigated thoroughly in local search. As initial steps in local search personalization, we propose a model for personalizing search results in a local search engine using a hybrid of profile- and click-based user modeling methods. User profiles are used to compare local search results to the topical interests of users and the specific businesses in which they have shown interest by way of search result “clicks”. Our model is tested through a user study and is shown to result in significantly improved mean average precision over the baseline ranking system.
Although experiments were conducted in the context of Local Search, the framework for modelling user interests is transferable to any domain where semantic similarity between users or user and objects is of interest. If you have any questions or comments about our approach, I’d be happy to discuss it in person or electronically – just give me a shout!
I would like to thank the staff at GenieKnows.com for their assistance and feedback both during the experiments and during writing the paper. Special shouts go to Stephanie Armsworthy, Jason Hines, and Jacek Wolkowicz, and Tapajyoti Das. Additional acknowledgements will appear in the paper.