Why did you record this video?An exploratory study on user intentions for videoproduction.Mathias Lux* & Jochen Huber§* Klagenfurt University, AT§ Technische Universität Darmstadt, DE This work is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 3.0
Motivation• Novelty of intentions in MMIS – have not (yet) been investigated thoroughly• Hard, interdisciplinary problem – fuzzy, social, deals with people• Intentions are diverse – have potential for distinguishing between different user groups (cc) by bitzcelt, http://www.flickr.com/photos/bitzcelt/
Goals• Find out if there is a taxonomy that can be used for MMIS.• Support or reject current approaches.• Find path towards a usable model supported by statistics. (cc) by jam343 http://www.flickr.com/photos/jam343
Roots• A taxonomy of web search (2002), A. Broder – navigational – “dublin wikipedia” – informational – “day tour dublin” – transactional – “book hotel in dublin”• Understanding user goals in web search (2004), D. Levinson & D. Rose – transactional -> resource – more fine-grained sub categories – informational > 60%
Roots• A classification scheme for user intentions in image search (2010), M. Lux, C. Kofler, O. Marques – 4th category: mental image – categories overlap
Photo production• The ubiquitous camera: An in-depth study of camera phone use (2005), T. Kindberg et al. – Affection vs. function – Social vs. individual
Video production• Video microblogging: your 12 seconds of fame (2010) N. Bornoe & L. Barkhuus – social collaboration (not individual) – self expression, entertainment, self representation• Practices in creating videos with mobile phones (2009), A. Puikkonen et al. – preserve moment of interest – sharing ”occasionally”, not by default
Methodology• Exploratory study – 20 participants (16m, 4f) – semi-structured interviews• Interviews – demographics & general usage – communication & recording habits• Instances – 48 situations were reported
Research QuestionsUsing Kindberg’s taxonomy as a basis• Are Kindberg’s classes disjoint? – are there instances that indicate overlap?• Is a 2D space sufficient to describe video production intentions – need for other dimensions? (cc) by oberazzi, http://www.flickr.com/photos/oberazzi/
Analysis• Clustering of instances – similar instances go together – grouped manually – discussed grouping – multiple assignments possible (cc) by alastanton, http://www.flickr.com/photos/alanstanton
Clustering• Preservation – Storing a scene to view it later• Sharing – Showing scenes to others• Affection – Capturing a scene due to emotion• Functional – Video is part of a job, hobby, etc.• Technical interest – E.g. trying out a camera• Other – Unknown or unmentioned intentions, etc.
Results• Nearly all of the videos (39 /48) were taken for sharing them. – 29 of the 39 instances: family, friends, colleagues, other closed groups.• Affection - 23 instances• Preservation - 19 instances
Results• Do class assignments co-occur?• Cross-tabulation – phi can be read like a correlation coefficient – -1 <= min <= phi <= max <= 1 – min, max due to different number of assignments
Discussion• Multiple assignments – 81% were assigned to more than 1 cluster – Are classes disjoint (e.g. function vs. affection)?P4 mentioned a video he took on a mountain while snowboarding. He recorded the video because he “took it because [he loves] snowboard video tricks and [he thinks] that it is very important to reconsider them to improve [his own] technique”.
DiscussionP10 reported “First my friend is so good at singing and also charming and second he was about to leave the city and that was our last meeting. So I took the video to remember the night”• Ad-hoc affection vs.• Preservation
Discussion• Preservation opposite of sharing? – No correlation in our data (A)• Function & preservation go together? – maximum neg. correlation (B) B A
Conclusion• Intention classes not disjoint in the domain of video production• Kindberg’s taxonomy is not sufficient for video production• Preservation, sharing, affection & function are 4 valid classes to start with.
Future work• Our proposed structure is biased by – the small data set – the convenience sample – the questions asked• Collected a data set for photos – 1,309 photos + intentions of their photographers – mturk validation and QA of the survey results• Collection of a video data set• Application in domains (cc) by thevince, http://www.flickr.com/photos/thevince
Thanks ….. for your interestmore on user intentions: http://tinyurl.com/mlux-iteccheck out LIRe CBIR library: http://www.semanticmetadata.net email@example.com (cc) http://www.jumpingbrain.org/