by Dong Nguyen, Thomas Demeester, Dolf Trieschnigg, and Djoerd Hiemstra
A publicly available dataset for federated search reflecting a real web environment has long been absent, making it difficult for researchers to test the validity of their federated search algorithms for the web setting. We present several experiments and analyses on resource selection on the web using a recently released test collection containing the results from more than a hundred real search engines, ranging from large general web search engines such as Google, Bing and Yahoo to small domain-specific engines.
First, we experiment with estimating the size of uncooperative search engines on the web using query based sampling and propose a new method using the ClueWeb09 dataset. We find the size estimates to be highly effective in resource selection. Second, we show that an optimized federated search system based on smaller web search engines can be an alternative to a system using large web search engines. Third, we provide an empirical comparison of several popular resource selection methods and find that these methods are not readily suitable for resource selection on the web. Challenges include the sparse resource descriptions and extremely skewed sizes of collections.
The slides of the CLEF keynote can be downloaded below
A case for search specialization and search delegation
Evaluation conferences like CLEF, TREC and NTCIR are important for the field, and keep being important because there is no “one-size-fits-all” for search engines. Different domains need different ranking approaches: For instance, Web search benefits from analyzing the link graph; Twitter search benefits from retweets and likes; Restaurant search benefits from geo-location and reviews; Advertisement search need bids and click-through, etc. Researching many domains will learn us more about the need and the value of the specialization of search engines, and about approaches that can quickly learn rankings for new domains using for instance learning-to-rank and clever feature selection.
A search engine that provides results from multiple domains, therefore better delegates its queries to specialized search engines. This brings up unique research questions on how to best select a specialized search engine. The TREC Federated Web Search track, that ran in 2013 and 2014, studied these questions in two tasks: the resource selection task studied how to select, given a query but before seeing the results for the query, the top specialized search engines for a query. The vertical selection task studied how to select the top domains from a predefined set of domains such as news, video, Q&A, etc.
I will present the lessons that we learned from running the Federated Web Search track, focusing on successful approaches to resource selection and vertical selection. I will conclude the talk by discussing our steps to take this work to full practice by running the University of Twente's search engine as a federation of more than 30 smaller search engines, including local databases with news, courses, publications, as well as results from social media like Twitter and YouTube. The engine that runs U. Twente search is called Searsia and is available as open source software at: http://searsia.org.
On December 06 and 07 2016 The Netherlands School for Information and Knowledge Systems (SIKS) and Statistics Netherlands (CBS) organize a two day tutorial on the management of Big Data, the DataCamp, hosted at the University of Twente.
The Data Camp's objective is to use big data sets to produce valuable and innovative answers to research questions with societal relevance. SIKS PhD students and CBS data analysts will learn about big data technologies and create, in small groups, feasibility studies for a research question of their choice.
Participants get access to predefined CBS research questions and massive datasets, including a large collection of Dutch Tweets, traffic data from Dutch high ways, and AIS data from ships. Participants will get access to the Twente Hadoop cluster, a 56 node cluster with almost 1 petabyte of storage space. The tutorial focuses on hands-on experience. The Data Camp participants will work in small, mixed teams in an informal setting, which stimulates intense contact with technologies and research questions. Experienced data scientists will support the teams by short lectures and hands-on support. Short lectures will introduce technologies to manage and visualize big data, that were first adopted by Google and are now used by many companies that manage large datasets. The tutorial teaches how to process terabytes of data on large clusters of commodity machines using new programming styles like MapReduce and Spark. The tutorial will be given in English and is part of the educational program for SIKS PhD students.
Also see the SIKS announcement.
by Mostafa Dehghani, Hosein Azarbonyad, Jaap Kamps, Djoerd Hiemstra, and Maarten Marx
Users tend to articulate their complex information needs in only a few keywords, making underspecified statements of request the main bottleneck for retrieval effectiveness. Taking advantage of feedback information is one of the best ways to enrich the query representation, but can also lead to loss of query focus and harm performance – in particular when the initial query retrieves only little relevant information – when overfitting to accidental features of the particular observed feedback documents. Inspired by the early work of Hans Peter Luhn, we propose significant words language models of feedback documents that capture all, and only, the significant shared terms from feedback documents. We adjust the weights of common terms that are already well explained by the document collection as well as the weight of rare terms that are only explained by specific feedback documents, which eventually results in having only the significant terms left in the feedback model.
Our main contributions are the following. First, we present significant words language models as the effective models capturing the essential terms and their probabilities. Second, we apply the resulting models to the relevance feedback task, and see a better performance over the state-of-the-art methods. Third, we see that the estimation method is remarkably robust making the models insensitive to noisy non-relevant terms in feedback documents. Our general observation is that the significant words language models more accurately capture relevance by excluding general terms and feedback document specific terms.
To be presented at the 25th ACM International Conference on Information and Knowledge Management (CIKM 2016) on October 24-28, 2016 in Indianapolis, United States.
by Suzan Verberne, Maya Sappelli, Djoerd Hiemstra, and Wessel Kraaij
We evaluate five term scoring methods for automatic term extraction on four different types of text collections: personal document collections, news articles, scientific articles and medical discharge summaries. Each collection has its own use case: author profiling, boolean query term suggestion, personalized query suggestion and patient query expansion. The methods for term scoring that have been proposed in the literature were designed with a specific goal in mind. However, it is as yet unclear how these methods perform on collections with characteristics different than what they were designed for, and which method is the most suitable for a given (new) collection. In a series of experiments, we evaluate, compare and analyse the output of six term scoring methods for the collections at hand. We found that the most important factors in the success of a term scoring method are the size of the collection and the importance of multi-word terms in the domain. Larger collections lead to better terms; all methods are hindered by small collection sizes (below 1000 words). The most flexible method for the extraction of single-word and multi-word terms is pointwise Kullback-Leibler divergence for informativeness and phraseness. Overall, we have shown that extracting relevant terms using unsupervised term scoring methods is possible in diverse use cases, and that the methods are applicable in more contexts than their original design purpose.
To appear in Information Retrieval.
by Jeroen Vonk
Within the field of Computer Science a lot of previous and current research is done on model checking. Model checking allows researchers to simulate a process or system, and exhaustively test for wanted or non-wanted properties. Logically, the result of these test are as dependable as your model represents the actual system. The best model then, would be a model representing the system down to its last atom, allowing for every possible interaction with the model. The model of course will become extremely large, a situation known as state space explosion. Current research therefore focuses on:
- Storing larger models
- Processing large models faster and smarter
- Reducing the size of models, whilst keeping the same properties
In this thesis we will focus on reducing the size of the models using bisimulation reduction. Bisimulation reduction allows to identify similar states that can be merged whilst preserving certain properties of the model. These similar, or redundant states will be identified by comparing them with other states in the model using a bisimulation relation. The bisimulation relation will identify states showing the same behavior, that therefore can be merged. This process is called bisimulation reduction. A common method to determine the smallest model is using partition refinement. In order to use the algorithm on large models it needs to be scalable. Therefore we will be using a framework for distributed processing that is part of Hadoop, called MapReduce. Using this framework provides us with a robust system that automatically recovers from e.g. hardware faults. The use of MapReduce also makes our algorithm scalable, and easily executed at third party clusters.
During our experiments we saw that the execution-time for a MapReduce job takes a relatively long time. We have estimated that there is a startup cost for each job of circa 30 seconds. This means that the reduction of transition systems that need a lot of iterations can be very high. Extreme cases such as the vasy 40 60 which take over 20,000 iterations therefore could not be benchmarked within an acceptable time-frame. Each iteration all of our data is passed over the disk. Therefore it is not unreasonable to see a factor 10-100 slow down compared to a mpi-based implementation (e.g. LTSmin). From our experiments we have concluded that the separate iteration times of our algorithm scale linearly up to 108 transitions for strong bisimulation and 107 for branching bisimulation. On larger models the iteration time increases exponentially, therefore we where not able to benchmark our largest model.
Beyond Movie Recommendations: Solving the Continuous Cold Start Problem in E-commerce Recommendations
by Julia Kiseleva, Alexander Tuzhilin, Jaap Kamps, Melanie Mueller, Lucas Bernardi, Chad Davis, Ivan Kovacek, Mats Stafseng Einarsen, Djoerd Hiemstra
Many e-commerce websites use recommender systems or personalized rankers to personalize search results based on their previous interactions. However, a large fraction of users has no prior interactions, making it impossible to use collaborative filtering or rely on user history for personalization. Even the most active users may visit only a few times a year and may have volatile needs or different personas, making their personal history a sparse and noisy signal at best. This paper investigates how, when we cannot rely on the user history, the large scale availability of other user interactions still allows us to build meaningful profiles from the contextual data and whether such contextual profiles are useful to customize the ranking, exemplified by data from a major online travel agent Booking.com.
Our main findings are threefold: First, we characterize the Continuous Cold Start Problem (CoCoS) from the viewpoint of typical e-commerce applications. Second, as explicit situational context is not available in typical real world applications, implicit cues from transaction logs used at scale can capture essential features of situational context. Third, contextual user profiles can be created offline, resulting in a set of smaller models compared to a single huge non-contextual model, making contextual ranking available with negligible CPU and memory footprint. Finally we conclude that, in an online A/B test on live users, our contextual ranker increased user engagement substantially over a non-contextual baseline, with click-through-rate (CTR) increased by 20%. This clearly demonstrates the value of contextual user profiles in a real world application.
by Marco Schultewolter
Often, software providers ask users to insert personal data in order to grant them the right to use their software. These companies want the user profile as correct as possible, but users sometimes tend to enter incorrect information. This thesis researches and discusses approaches to automatically verify this information using third-party web resources.
Therefore, a series of experiments is done. One experiment compares different similarity measures in the context of a German phone book directory for again different search approaches. Another experiment takes the approach to use a search engine without a specific predefined data source. Ways of finding persons in search engines and of extracting address information from unknown websites are compared in order to do so.
It is shown, that automatic verification can be done to some extent. The verification of name and address data using external web resources can support the decision with Jaro-Winkler as similarity measure, but it is still not solid enough to only rely on it. Extracting address information from unknown pages is very reliable when using a sophisticated regular expression. Finding persons on the internet should be done by using just the full name without any additions.
Journal Citation Statistics for Library Collections using Document Reference Extraction Techniques
by Steven Verkuil
Providing access to journals often comes with a considerable subscription fee for universities. It is not always clear how these journal subscriptions actually contribute to ongoing research. This thesis provides a multistage process for evaluating which journals are actively referenced in publications. Our software tool for journal citation reports, CiteRep, is designed to aid decision making processes by providing statistics about the number of times a journal is referenced in a document set. Citation reports are automatically generated from online repositories containing PDF documents. The process of extracting citations and identifying journals is user and maintenance friendly. CiteRep allows to filter generated reports by year, faculty and study providing detailed insight in journal usage for specific user groups. Our software tool achieves an overall weighted precision and recall of 66,2% when identifying journals in a fresh set of PDF documents. While leaving open some areas of improvement, CiteRep outperforms the two most popular citation parsing libraries, ParsCit and FreeCite with respect to journal identification accuracy. CiteRep should be considered for creation of journal citation reports from document repositories.
Clone CiteRep on Github.
by Mohammadreza Khelghati
Data is one of the keys to success. Whether you are a fraud detection officer in a tax office, a data journalist or a business analyst, your primary concern is to access all the relevant data to your topics of interest. In such an information-thirsty environment, accessing every source of information is valuable. This emphasizes the role of the web as one of the biggest and main sources of data. In accessing web data through either general search engines or direct querying of deep web sources, the laborious work of querying, navigating results, downloading, storing and tracking data changes is a burden on shoulders of users. To decrease this intensive labor work of accessing data, (semi-)automatic harvesters have a crucial role. However, they lack a number of functionalities that we discuss and address in this work.
In this thesis, we investigate the path towards a focused web harvesting approach which can automatically and efficiently query websites, navigate through results, download data, store it and track data changes over time. Such an approach can also facilitate users to access a complete collection of relevant data to their topics of interest and monitor it over time. To realize such a harvester, we focus on the following obstacles. First, we try to find methods that can achieve the best coverage in harvesting data for a topic. Although using a fully automatic general harvester facilitates accessing web data, it is not a complete solution to collect a thorough data coverage on a given topic. Some search engines, in both surface web and deep web, restrict the number of requests from a user or limit the number of returned results presented to him. We suggest an efficient approach which can pass these limitations and achieve a complete data coverage.
Second, we investigate reducing the cost of harvesting a website regarding the number of submitted requests by estimating its actual size. Harvesting tasks continue till they face the posed query submission limitations by search engines or consume all the allocated resources. To prevent this undesirable situation, we need to know the size of the targeted source. For a website that hides the true size of its residing data, we suggest an accurate method to estimate its size.
As the third challenge, we focus on monitoring data changes over time in web data repositories. This information is helpful in providing the most up-to-date answers to information needs of users. The fast evolving web adds extra challenges for having an up-to-date data collection. Considering the costly process of harvesting, it is important to find methods which facilitate efficient re-harvesting processes.
Lastly, we combine our experiences in harvesting with the studies in the literature to suggest a general designing and developing framework for a web harvester. It is important to know how to configure harvesters so that they can be applied to different websites, domains and settings.
These steps bring further improvements to data coverage and monitoring functionalities of web harvesters and can help users such as journalists, business analysts, organizations and governments to reach the data they need without requiring extreme software and hardware facilities. With this thesis, we hope to have contributed to the goal of focused web harvesting and monitoring topics over time.