Optimizing Search Interactions within Professional Social Networks Nikita Spirin Department of Computer Science, University of Illinois at Urbana-Champaign Thomas M. Siebel Center for Computer Science, 201 N Goodwin Ave, Urbana, IL 61801, USA
[email protected] ABSTRACT To help users cope with the scale and influx of new information, professional social networks (PSNs) provide a search functionality. However, most of the search engines within PSNs today only support keyword queries and basic faceted search capabilities overlooking serendipitous network exploration and search for relationships between entities. This results in siloed information and a limited search space. My thesis is that we must redesign all major elements of a search user interface, such as input, control, and informational, to enable more effective search interactions within PSNs. I will introduce new insights and algorithms supporting the thesis.
Keywords Search User Interaction; Query Log Analysis; Online Social Network; Job Search; Entity Search; Snippet; Filtering
Categories and Subject Descriptors H.3.3 [Information Search and Retrieval]: Information filtering, Retrieval models, Search process, Selection process
1.
KEY CONTRIBUTIONS
A large scale analysis of Facebook Graph Search (FBGS) query logs [3]. I examined how search behavior differs for different query types, how it changes for users from various demographic groups, and how it depends on the graph distance between a searcher and a person to be searched. The analysis revealed new findings and suggested many design implications: users search more for friends using NEQs and for non-friends using SQs, which shows their complementary roles in enabling effective PSN search; structured query (SQ) usage behavior has a wider variation across different demographics, and, hence, it makes sense to direct search personalization efforts on SQs. The crux of this work is that users do benefit from more control and highly interactive query suggestions — they engage in a new type of search behavior (exploratory search for non-friends). Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for third-party components of this work must be honored. For all other uses, contact the owner/author(s). WSDM 2016 February 22-25, 2016, San Francisco, CA, USA c 2016 Copyright held by the owner/author(s).
ACM ISBN 978-1-4503-3716-8/16/02. DOI: http://dx.doi.org/10.1145/2835776.2855092
An algorithm for relevance-aware search results filtering [2], which improves the quality of search results sorted by an attribute value (e.g. likes, date, salary). It is based on the dynamic programming, directly optimizes a given search quality metric, and, therefore, theoretically optimal. According to the experiments, it does outperform all baselines and lead to 2-4% increase in search quality. An algorithm for snippets generation for job search. It allows to generate informative snippets that contain the key information (resp. and req. for a job) right on a SERP and help users make clicks mostly on relevant results. Plus, it helps optimize content for mobile devices and avoid irregularities of job postings coming from multiple websites by converting them into a standardized structured representation. The algorithm leverages the power of big data to minimize supervision required for model training, and, hence, could be easily deployed internationally. Basedon a series of offline and online experiments: (1) the algorithm achieves high extraction accuracy (86% precision at 94% coverage for English [1] and 97% precision at 100% coverage by a highlytuned enterprise-grade model for Russian); (2) structured snippets improve the majority of user-centric search quality metrics (e.g. apply rate conditioned on click is 13% and total applies 1.6% up; number of short clicks by 5.5% down). A user study analyzing the effectiveness of deltasnippets for entity search. The common belief is that query-biased snippets are the most effective. However, it is not necessarily the case for entity search since there is less need to duplicate the information from a query in a snippet (all results exactly match the query). By reasoning from basic principles, I introduce the concept of delta-snippet, which shows information complementary to the query and more effectively uses a screen space (esp. on mobile), and investigate it via a user study. I examine how users interact with delta-snippets and how it changes users’ productivity.
2.
REFERENCES
[1] N. Spirin and K. Karahalios. Unsupervised approach to generate informative structured snippets for job search engines. WWW ’13. [2] N. Spirin, M. Kuznetsov, J. Kiseleva, Y. Spirin, and P. Izhutov. Relevance-aware filtering of tuples sorted by an attribute value via direct optimization of search quality metrics. SIGIR’15. [3] N. V. Spirin, J. He, M. Develin, K. G. Karahalios, and M. Boucher. People search within an online social network: Large scale analysis of facebook graph search query logs. CIKM ’14.
