The Search Solutions Conference which took place on 27 November 2013 promoted quality and diversity in the selection of speakers and topics – this quality driven choice of heterogenic elements was morphed into a homogeneous programme covering aspects from user-oriented search to enterprise-related search.
The conference provided attendees the abstract layers of dogmatic search topics that are of great interest in current industry and academia. The quality of the conference was reflected as well in the organization – a balanced event in terms of audience-size which created a favorable environment for an interaction process based on the exchange of knowledge.
The first session of the event, “The changing face of Search”, started with a presentation given by Behshad Behzadi from Google Switzerland on “Conversational Search”. The first part of the talk focused on illustrating the key elements of what the presenter considers to be the search of the future: elements such as the ability of the systems to answer questions, to anticipate the information need of the user, and last but not least, the focus of this presentation, the ability to search for information in a conversational style. The beauty of conversational search is that it offers human-computer interaction in terms of dialogue in a natural an intuitive way. There are of course major challenges that need to be considered such as understanding the user intent, the context of the discussion, the human language with its ambiguity or dealing with issues such as accents or background noise. More than that, all these challenges need to be addressed as much as possible from a generic perspective, independent of language – which introduces another big challenge. The second part of the presentation consisted in a live demonstration involving the presenter engaging in a conversation with the Google search engine. The purpose of the demonstration was to illustrate in an abstract manner how Google addressed the above mentioned challenges. The second presenter of the session, Wim Nijmeijer from Coveo, introduced the audience into enterprise search and talked about a product used to ensure the integration of multiple repositories in a homogeneous manner such that search can be done effortlessly across the whole enterprise data.
The last presenter of this first session was Nikolaos Nanas from NOOWIT who gave a talk on both the back-end and front-end of an application which aims to deliver relevant content to users. From a back-end perspective, the researcher presented the benefits of Nootropia, a biologically inspired model, over traditional IR approaches such as the Vector Space Model. Each piece of content is assigned a relevance code based on the interests of the users which are dynamic. The application is capable of ranking a piece of content even before it received any interaction from the community of users by comparing its characteristics against existing similar articles. From a front-end perspective, the researcher discussed how the application divides the screen space to display the relevant articles in such a manner that the most relevant pieces of information occupy the most space.
In the second session of the Search Solutions conference, Henning Rode from TextKernel provided us with a glimpse into the interface and indexing of Curriculum Vitae (CV) based repositories: exploiting rich search interfaces with guided queries via suggestive fields and bi-gram style search; additionally demonstrating how heuristics analysis of Job adverts can be used to encourage a cohesive coupling between applicant and role. With the digitalisation of multimedia content becoming more prolific, repositories such as the Digital Repository of Ireland have begun to appear. Sandra Collins and Dermot Frost provided an enlightening talk into the structure of such repositories, utilising current infrastructure technologies among Hydra, Ceph and Apache Solr. To search they lean on Solr, utilising meta-data for multimedia content.
The third session of the day, “Understanding the User”, started with a talk given by Filip Radlinski from Microsoft Research on “Sensitive Online Search Evaluation”.
The Microsoft researcher expressed the difference between the offline and online evaluation methods and focused his discussion on the latter. Useful insights into the variety of evaluation objectives such as determining the correctness of a result set or its practicability were provided. Among other mentioned aspects, the talk included the fact that a good way to assess the relevance of the results retrieved by a search engine is to analyze the behavior of the users (users’ fidelity) and the fact that the correct evaluation of the results is highly dependent on choosing the right type of metrics (absolute vs. relative metrics). The researcher also illustrates how interleaving is used for online evaluation to test different ranking algorithms – the result of this type of evaluation can only offer information on whether one ranking is better than the other, but nothing more than that (e.g.: cannot state whether ranking A outperforms ranking B with x%). The second talk of this third session was given by Mounia Lalmas from Yahoo! Labs and it focused on the understanding of the user engagement in the context of a search engine. The user engagement is viewed by the presenter as an aspect of user experience and it involves things such as aesthetics – a good user interface, novelty – presenting not only relevant results but also interesting ones, or reputation. The focus of the presenter’s research is on analyzing relatively short time interests in the user engagement. This session was concluded by Margaret Hanley from Transform UK who delivered a talk on the elements that allow one to understand the search intent of the users. The core challenge in this area is represented by the “short length” characteristic of search queries (e.g.: 1, 2 words) – moving outside the individual query level, one can extract useful insights for understanding intent in search by analyzing patterns in the historical data and by performing various types of experiments such as user testing, query review or understanding the content within the search domain.
Enterprise Search was the focus of Session 4. Helen Lippell from Press Association opened this last session of the day with a presentation on the problems companies have to deal with at the “intranet search” level. The core message expressed was that search is not an island; it is part of an information ecosystem and the organizations which successfully address the search challenges are the ones which handle effectively both the technology and the data aspects. Peter Wallqvist from Ravn Systems moved the discussion into the corporate social networking area. The delivered presentation focused on the benefits brought by the symbiotic relationship between enterprise search and social networking – the search results retrieved in this context consider the position of the search entity in the Knowledge Graph (a graph of interlinked people and interests from the organization) in relation to other candidate results and favor the newly introduced content. The presenter outlined at the same time the main limitations of universal search systems: little distinction between quality content, limited ways to control ranking and relevance or the inability to have access to the newly introduced content until the search operation is performed again. The final talk of the day was delivered by Kristian Norling from Findwise. This last presentation was a reflection of a series of Enterprise Search related survey results from 2013. According to a statistic offered by the presenter, 78% of the survey participants consider Enterprise Search a critical success factor, but what is really surprising is that most companies do not have even 1 person trained and dedicated entirely to search.
The final fishbowl session allowed for group interaction. This session focused primarily on the networking undertone of the conference — the interactions between Industry and Academia. Key topics included the possibility of allowing industrial experience to research students, to not only facilitate the transfer of knowledge from research to industrial application, but also allow industry to guide and hone research to current requirements. A hot topic among researchers is accessibility of data, often great ideas are stifled by the limitations of public data; this topic although came to no direct conclusion, it reminded both sides of the importance of keeping communication channels open and the benefits of conferences such as Search Solutions.
Silviu Paun – University of Essex
Stuart James – University of Surrey