26
Rank all the things! @jsuchal @SynopsiTV

Ján Suchal - Rank all the things!

Embed Size (px)

DESCRIPTION

Google search results and Netflix recommendations share more in common that you might think at first look. Let’s take a deep dive into recent research and see what makes search results or recommendations good and why. We’ll talk about how users interact with search or recommendations and what are the key metrics we can measure and improve. Make no mistake, this concerns you even if you never planned to build a search or recommendation engine.

Citation preview

Page 1: Ján Suchal - Rank all the things!

Rank all the things!@jsuchal@SynopsiTV

Page 2: Ján Suchal - Rank all the things!
Page 3: Ján Suchal - Rank all the things!

Blogs, newsletters

How do you learn things?

Courses, training

Conferences Work

Page 4: Ján Suchal - Rank all the things!

Research papers?

Page 5: Ján Suchal - Rank all the things!

WHY NOT?

Page 6: Ján Suchal - Rank all the things!

WHY NOT?

“It’s not useful for the real-world.”

“I wouldn’t understand any of

that.”

Page 7: Ján Suchal - Rank all the things!

About me

PhD dropout FIIT STU Bratislava

foaf.sk, otvorenezmluvy.sk, govdata.sk

sme.sk news recommender

developer @ SynopsiTV

Page 8: Ján Suchal - Rank all the things!

My workflow

Page 9: Ján Suchal - Rank all the things!

My workflow

MAGIC!M

AG

IC!

MA

GIC

!

Page 10: Ján Suchal - Rank all the things!

Search vs. recommender engine

Search engine

input: queryoutput: list of results

Recommendation engine

input: movieoutput: list of similar movies

Page 11: Ján Suchal - Rank all the things!

Academic Mode

Page 12: Ján Suchal - Rank all the things!

Accurately interpreting clickthrough data as implicit feedback

Thorsten Joachims, Laura Granka, Bing Pan, Helene Hembrooke, and Geri Gay. Accurately interpreting clickthrough data as implicit feedback. In Proceedings of the 28th annual international ACM SIGIR conference on Research and development in Information retrieval, SIGIR ’05, pages 154–161, New York, NY, USA, 2005. ACM.

Significant on two-tailed tests at a 95% confidence level !!!

Page 13: Ján Suchal - Rank all the things!

Accurately interpreting clickthrough data as implicit feedback

Thorsten Joachims, Laura Granka, Bing Pan, Helene Hembrooke, and Geri Gay. Accurately interpreting clickthrough data as implicit feedback. In Proceedings of the 28th annual international ACM SIGIR conference on Research and development in Information retrieval, SIGIR ’05, pages 154–161, New York, NY, USA, 2005. ACM.

Page 14: Ján Suchal - Rank all the things!

Accurately interpreting clickthrough data as implicit feedback

Page 15: Ján Suchal - Rank all the things!

Evaluation Metrics

● Mean Average Precision @ N○ probability of target result being in top N items

● Mean Reciprocal Rank○ 1 / rank of target result

● Normalized Discounted Cumulative Gain● Expected Reciprocal Rank

Page 16: Ján Suchal - Rank all the things!

Optimizing search engines using clickthrough data

Thorsten Joachims. Optimizing search engines using clickthrough data. In Proceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data mining, KDD ’02, pages 133–142, New York, NY, USA, 2002. ACM.

Page 17: Ján Suchal - Rank all the things!

Optimizing search engines using clickthrough data

Page 18: Ján Suchal - Rank all the things!

Query chains: learning to rank from implicit feedback

Filip Radlinski and Thorsten Joachims. Query chains: learning to rank from implicit feedback. In KDD ’05: Proceeding of the eleventh ACM SIGKDD international conference on Knowledge discovery in data mining, pages 239–248, New York, NY, USA, 2005. ACM.

Page 19: Ján Suchal - Rank all the things!

On Caption Bias in Interleaving Experiments

Katja Hofmann, Fritz Behr, and Filip Radlinski: On Caption Bias in Interleaving Experiments In Proceedings of the ACM Conference on Information and Knowledge Management (CIKM) 2012

Page 20: Ján Suchal - Rank all the things!

On Caption Bias in Interleaving Experiments

Page 21: Ján Suchal - Rank all the things!

Fighting Search Engine Amnesia: Reranking Repeated Results

Milad Shokouhi, Ryen W. White, Paul Bennett, and Filip Radlinski. Fighting search engine amnesia: reranking repeated results. In Proceedings of the 36th international ACM SIGIR conference on Research and development in information retrieval, SIGIR ’13, pages 273–282, New York, NY, USA, 2013. ACM.

In this paper, we observed that the same results are often shown to users multiple times during search sessions. We showed that there are a number of effects at play, which can be leveraged to improve information retrieval performance. In particular, previously skipped results are much less likely to be clicked, and previously clicked results may or may not be re-clicked depending on other factors of the session.

Page 22: Ján Suchal - Rank all the things!

Challenges

Page 23: Ján Suchal - Rank all the things!

Diversification

Page 24: Ján Suchal - Rank all the things!

Group recommendations

Page 25: Ján Suchal - Rank all the things!

Context-aware recommendations

Time of day

DeviceMood

Season

Location

Page 26: Ján Suchal - Rank all the things!

Seriousrecommenders and search?Get in touch!

@synopsitv @jsuchal