Lubos Steskal works as a Data Scientist in TV 2. After getting his PhD in computer science, he participated in a research project that set out to map out the structure and discourse of the climate change blogosphere. He also worked in a news and media recommendation and aggregation start-up, a bank and is currently enjoying himself in TV 2. In his spare time, he co-founded and later sold an escape room company.

Exploring the many uses of embedding based textual similarity
I will be showing how representing text/words as relatively low dimensional vecotrs (300 dimensions) "latent semantic" space allows for similarity identification beyond the matching of exact words and how this can be used to aid in the cold start problem for recommandation as well as grouping related content, generating document tags, and matching content across all our content domains (streaming, TV news channel, tv2.no content)