Diversity-Enhancing Professional Social Matching
Goals: social matching in knowledge work
- Enable unexpected social encounters in professional life
- ... without amplifying the emergence of echo chambers (homophily, triadic closure)
- ... by helping users avoid human bias in decision-making and identifying optimal similarity-diversity between matched people
Design requirements: context-sensitivity, systemic perspective, user-system cooperation, proactive persuation
(Olsson et al., CACM manuscript)
Our approach
-
Big Social Data (Olshannikova et al., 2017) on people, their networks and interactions as the fuel for the digital artefacts
-
Mapping existing social networks and identifying new potential connections between existing clusters
-
Action Design Research (Sein et al. 2011)
-
User experiments & field studies
-
Bridging the methodological gap
between disciplines!
Theoretical background
Social structure of organizations
Social structure core concepts
- Weak ties (Granovetter, 1973)
- Homophily (McPherson, Smith-lovin & Cook, 2001)
- Triadic closure (Granovetter, 1973)
- Echo chambers (Sunstein, 2009)
- Cyberbalkanization (Van Alstyne & Brynjolfsson, 1996)
- Filter bubbles (Pariser, 2011)
- Structural holes (Burt, 2001)
Social structure in knowledge work
- Weak ties are conduits of novel information (Aral, 2016)
- Brokering structural holes is beneficial for individuals (Burt, 2004)
- Access to heterogeneous knowledgde drives innovation performance (Rodan & Galunic, 2004)
- Diversity-bandwidth tradeoff is needed to navigate strong and weak ties (Aral & Van Alstyne, 2011)
Experiment
- Bibliographic data from dblp: 3M articles (title, authors, venue)
- Feature vectors for each author: words in article titles (bigrams, TF-IDF)
- Similarity is calculated between all pairs of actors (cosine similarity)
- Recommendations: Most similar, somewhat similar, not similar
Experiment #1 (submitted to CSCW)
Echo chamber hypothesis:
- Actors prefer to connect with similar actors (homophily)
- New connections are formed between second-tier connections or friends-of-friends (triadic closure)
- Is the simplified similarity-based recommendation strategy able to produce new potential connections for actors?
- Does the similarity between actors indeed predict the perceived relevance of a connection?
Experiment #2
Under development
Experiment #2 (under development)
- Breaking echo chambers:
1. Using a network view of actors around an institution
2. Social network provides context for recommendations
- Is the full network view valuable to users in exploring the social space?
- Do the users prefer to connect with actors in their second-tier network?
- Are the users willing to move beyond second-tier connections?
Experiment #2 details
- Twitter followers of the three universities in Tampere to be merged into Tampere University
- 300 most recent tweets from each follower
- Mention-based social network
- Topic modeling (LDA)-based interest vectors
- Recommendation strategies under development and to be discussed
- User experiment to be run in the next few months
Toward systemic serendipity
Serendipitous social encounters
Title
- Content
We adapt and extend the McCay-Peet & Toms (2015) serendipity model to social serendipity (submitted to CSCW 2018)
Outlook
Where should we go from here?
Social recommender systems
Concluding remarks
- We take a social network-centric viewpoint to social matching and seek to increase diversity by identifying potential weak ties into organizations
- We subscribe to Tsai and Brusilovsky (2018) in that instead of relevance-first approach, we should develop ways to increase transparency and user control in social matching
- However, nudging-based systems will also be developed and we should be able to inform their design
- Key short term objective is to investigate the perceived value of full social network-approach in identifying and recommending connections