The Search Interaction Optimization Toolkit – The Essence of my PhD Thesis

SIO Toolkit Logo
Logo of the SIO Toolkit.

My PhD thesis introduces a novel methodology that is named Search Interaction Optimization (SIO) and is used for designing, evaluating and optimizing search engine results pages (so-called SERPs). As a proof-of-concept of this new methodology, I’ve developed a corresponding SIO toolkit, which comprises a total of seven components1 (most of which have already been introduced in previous posts):

  1. Inuit, a new instrument for usability evalutation;
  2. WaPPU, a tool for Usability-based Split Testing;
  3. a catalog of best practices for creating better usable SERPs, which together with WaPPU and a special add-on forms
  4. S.O.S., a tool for automatically evaluating and optimizing SERPs;
  5. TellMyRelevance! (TMR), a novel pipeline that predicts the relevance of search results from client-side interactions;
  6. StreamMyRelevance! (SMR), a streaming-based version of TMR that works in real-time rather than batch-wise; and
  7. a set of requirements for current & future search interfaces, which has been derived from an empirical study with German-speaking users.
SIO Methodology Logo
Logo of the SIO Methodology.

Describing the design and development of the above components and evaluating their effectiveness and feasibility makes for a major part of my thesis. Now, I’ve finally managed to organize all of them in terms of GitHub repos2, which I make available through a new website I have specifically created for my PhD project: http://www.maxspeicher.com/phdthesis/. In particular, on that site you can filter the components depending on whether you want to design, evaluate and/or optimize a SERP. It also lists all of the related publications including links to the corresponding full texts (via ResearchGate). In case you are actually interested in all that fancy research stuff3—have fun browsing, reading & playing around! 🙂

1 The logo of the SIO toolkit features only six tiles because S.O.S. and the catalog of best practices are treated as one component there.
2 Because my PhD project was carried out in cooperation with Unister GmbH (Leipzig), unfortunately it’s not possible for me to provide the source codes of all components via GitHub, as some contain company secrets.
3 Which I doubt. 😉

Advertisements

StreamMyRelevance! Predicting Search Result Relevance from Streams of Interactions

SMR paper @ ICWE2014Guessing the relevance of delivered search results is one of the biggest issues for today’s search engines. The particular problem is that it’s difficult to obtain explicit statements from users about whether they found what they were searching for. Clicks are commonly used to guess relevance (using so-called “click models”) but they are far from being a perfect indicator. Particularly, a user might click a search result, but then return to the results page because the visited webpage was useless. Also, it’s possible that no clicks happen at all if the desired piece of information is already shown on the results page (e.g., in terms of an info box).

To tackle the above shortcoming, we have investigated the suitability of implicit feedback in terms of mouse cursor interactions for predicting the relevance of search results. For this, we developed StreamMyRelevance!—a system that receives streams of interactions and relevance judgments and trains statistical models from these in near real-time. The models can then be used to infer relevance from interactions in the future. The relevance judgments we’re using to train our models can either be implicit (e.g., a completed booking process in the case of hotel search) or explicit (e.g., statements by paid quality raters/crowdworkers).

Analysis of a large amount of real-world interaction data from two e-commerce portals showed that StreamMyRelevance! is able to train good models that show the tendency to perform better than a state-of-the-art click model solution1 that is successfully used in industry. Our results particularly underpin the benefit of using interaction data other than clicks for guessing the relevance of search results.

We have summarized the design and evaluation of our system in a full research paper2 that will be presented at the 2014 International Conference on Web Engineering (ICWE). The conference proceedings will be published by Springer and the final version of our paper will be available at link.springer.com. Special thanks go to Sebastian Nuck, who helped with development and evaluation of StreamMyRelevance! in the context of his Master’s Thesis at Leipzig University of Applied Sciences.

1 Chao Liu, Fan Guo, and Christos Faloutsos (2009). “BBM: Bayesian Browsing Model from Petabyte-Scale Data”. In Proc. KDD.
2 Maximilian Speicher, Sebastian Nuck, Andreas Both and Martin Gaedke (2014). “StreamMyRelevance! Prediction of Result Relevance from Real-Time Interactions and its Application to Hotel Search”. In Proc. ICWE.

4 Submissions accepted at International Conference on Web Engineering (ICWE)

End of February, I submitted four contributions to the 14th International Conference on Web Engineering: two full papers, one demo and one poster. Of these four submissions, all were accepted and will be presented at the conference, which is to be held in Toulouse (see map below) from July 1 to July 4. In the following, I’ll give a quick overview of the accepted papers. A more detailed explanation of my current research will be the subject of one or two separate articles.

  • Maximilian Speicher, Sebastian Nuck, Andreas Both, Martin Gaedke: “StreamMyRelevance! Prediction of Result Relevance from Real-Time Interactions and its Application to Hotel Search” — This full paper is based on Sebastian Nuck’s Master thesis. He developed a system for processing user interactions collected on search results pages in real-time and predicting the relevance of individual search results from these.
  • Maximilian Speicher, Andreas Both, Martin Gaedke: “Ensuring Web Interface Quality through Usability-based Split Testing” — This full paper proposes a new approach to split testing that is based on the actual usability of the investigated web interface rather than pure conversion maximization. We have trained models for predicting usability from user interactions and from these have also derived additional interaction-based heuristics for comparing search results pages.
  • Maximilian Speicher, Andreas Both, Martin Gaedke: “WaPPU: Usability-based A/B Testing” — This demo accompanies our paper about Usability-based Split Testing. The WaPPU tool builds upon this new concept and demonstrates how usability can be predicted from user interactions using automatically learned models.
  • Maximilian Speicher: “Paving the Path to Content-centric and Device-agnostic Web Design” — This poster is based on one of my previous posts. It provides a review of motherfuckingwebsite.com, which satirically claims to be a perfect website. Based on current research, we suggest improvements to the site that follow a strictly content-centric and device-agnostic approach.

My PhD research is supervised by Prof. Dr.-Ing. Martin Gaedke (VSR Research Group, Chemnitz U of Technology) and Dr. Andreas Both (R&D, Unister GmbH) and funded by the ESF and the Free State of Saxony.

2013 in Review: Search Interaction Optimization

On January 1, 2013 I started my PhD studies at Chemnitz University of Technology in cooperation with Unister GmbH, Leipzig. My project is about automatic methods for optimizing search engine results pages (e.g., http://www.google.com/#q=Hello%2C+World!) with respect to result quality and interface usability. The official working title is Search Interaction Optimization: A Design Thinking Approach.

During my first year as a PhD student, I have published three papers (as first author). A full paper about my first milestone has been presented at the International Conference on Knowledge and Information Management (CIKM), which was held in San Francisco in October/November.1 This milestone was about deducing the relevance of search results from user interactions on the search engine results page. Using large amounts of anonymous interaction data from two real-world hotel booking portals, we could show that it is possible to learn according relevance models of reasonable quality.

A second (short) paper was presented at the PhD Symposium of the International Conference on Web Engineering (ICWE) in July.2 The paper addressed the attempt of learning a common model that predicts usability based on training data from a group of similar webpages (e.g., online news articles). However, we concluded that this is not easily possible, because differences in low-level page structure and user intention counter model precision. Thus, additional preprocessing steps are necessary to minimze these influences. The usability evaluation described in this paper was based on a novel instrument for measuring usability whose items have been specifically designed for correlation with client-side interaction features. This Interface Usability Instrument (Inuit) was presented at the workshop “Methodological Approaches to HCI” in September.3 The above described is part of the second milestone of my PhD project, which is about automatic methods for optimizing interface usability. This milestone is my current work-in-progress and will be finished in 2014.

Alright, so much for my research in 2013. I’ll keep you updated with more fine-grained results during the new year.

1 Speicher, Both, Gaedke: “TellMyRelevance! Predicting the Relevance of Web Search Results from Cursor Interactions” (http://doi.acm.org/10.1145/2505515.2505703).
2 Speicher, Both, Gaedke: “Was that Webpage Pleasant to Use? Predicting Usability Quantitatively from Interactions” (http://dx.doi.org/10.1007/978-3-319-04244-2_33).
3 Speicher, Both, Gaedke: “Towards Metric-based Usability Evaluation of Online Web Interfaces” (http://dl.mensch-und-computer.de/handle/123456789/3399).