Today’s search engines allow users to discover relevant information in different types of modalities or media, e.g., web pages, text documents, images, or videos. It is, however, a challenging task to present mixed-modality result lists in an effective and easy-to-skim form. The two most commonly used approaches are to present the modalities side-by-side, each in a separate column of the result page; or to separate the modalities into multiple tabs. However, the field lacks a structured investigation on how the column or tab layout influence the users’ perception and usage of multimodal resources in an academic search task. In this paper, we present a user study (N=50) where the participants were asked to accomplish a search task for a fictive computer science seminar at the university. We evaluate the influence of the different layouts on (1) user search behavior (e.g., time until first resource is saved) and (2) the relevance of the selected resources for the task at hand. Finally, we discuss the results and possible implications for the design of multimodal search result presentation.

The emerging research field Search as Learning (SAL) investigates how the Web facilitates learning through modern information retrieval systems. SAL research requires significant amounts of data that capture both search behavior of users and their acquired knowledge in order to obtain conclusive insights or train supervised machine learning models. However, the creation of such datasets is costly and requires interdisciplinary efforts in order to design studies and capture a wide range of features. In this paper, we address this issue and introduce an extensive dataset based on a user study, in which 114 participants were asked to learn about the formation of lightning and thunder. Participants’ knowledge states were measured before and after Web search through multiple-choice questionnaires and essay-based free recall tasks. To enable future research in SAL-related tasks we recorded a plethora of features and person-related attributes. Besides the screen recordings, visited Web pages, and detailed browsing histories, a large number of behavioral features and resource features were monitored. We underline the usefulness of the dataset by describing three, already published, use cases.

Whole genome bisulfite sequencing is currently at the forefront of epigenetic analysis, facilitating the nucleotide-level resolution of 5-methylcytosine (5mC) on a genome-wide scale. Specialized software have been developed to accommodate the unique difficulties in aligning such sequencing reads to a given reference, building on the knowledge acquired from model organisms such as human, or Arabidopsis thaliana. As the field of epigenetics expands its purview to non-model plant species, new challenges arise which bring into question the suitability of previously established tools. Herein, nine short-read aligners are evaluated: Bismark, BS-Seeker2, BSMAP, BWA-meth, ERNE-BS5, GEM3, GSNAP, Last and segemehl. Precision-recall of simulated alignments, in comparison to real sequencing data obtained from three natural accessions, reveals on-balance that BWA-meth and BSMAP are able to make the best use of the data during mapping. The influence of difficult-to-map regions, characterized by deviations in sequencing depth over repeat annotations, is evaluated in terms of the mean absolute deviation of the resulting methylation calls in comparison to a realistic methylome. Downstream methylation analysis is responsive to the handling of multi-mapping reads relative to mapping quality (MAPQ), and potentially susceptible to bias arising from the increased sequence complexity of densely methylated reads.

Two modalities are often used to convey information in a complementary and beneficial manner, e.g., in online news, videos, educational resources, or scientific publications. The automatic understanding of semantic correlations between text and associated images as well as their interplay has a great potential for enhanced multimodal web search and recommender systems. However, automatic understanding of multimodal information is still an unsolved research problem. Recent approaches such as image captioning focus on precisely describing visual content and translating it to text, but typically address neither semantic interpretations nor the specific role or purpose of an image-text constellation. In this paper, we go beyond previous work and investigate, inspired by research in visual communication, useful semantic image-text relations for multimodal information retrieval. We derive a categorization of eight semantic image-text classes (e.g., "illustration" or "anchorage") and show how they can systematically be characterized by a set of three metrics: cross-modal mutual information, semantic correlation, and the status relation of image and text. Furthermore, we present a deep learning system to predict these classes by utilizing multimodal embeddings. To obtain a sufficiently large amount of training data, we have automatically collected and augmented data from a variety of datasets and web resources, which enables future research on this topic. Experimental results on a demanding test set demonstrate the feasibility of the approach.

Some recent approaches for character identification in movies and TV broadcasts are realized in a semi-supervised manner by assigning transcripts and/or subtitles to the speakers. However, the labels obtained in this way achieve only an accuracy of $80\% - 90\%$ and the number of training examples for the different actors is unevenly distributed. In this paper, we propose a novel approach for person identification in video by correcting and extending the training data with reliable predictions to reduce the number of annotation errors. Furthermore, the intra-class diversity of rarely speaking characters is enhanced. To address the imbalance of training data per person, we suggest two complementary prediction scores. These scores are also used to recognize whether or not a face track belongs to a (supporting) character whose identity does not appear in the transcript etc. Experimental results demonstrate the feasibility of the proposed approach, outperforming the current state of the art.

MOTIVATION: Next-generation sequencing has become an important tool in molecular biology. Various protocols to investigate genomic, transcriptomic and epigenomic features across virtually all species and tissues have been devised. For most of these experiments, one of the first crucial steps of bioinformatic analysis is the mapping of reads to reference genomes. RESULTS: Here, we present thorough benchmarks of our read aligner segemehl in comparison with other state-of-the-art methods. Furthermore, we introduce the tool lack to rescue unmapped RNA-seq reads which works in conjunction with segemehl and many other frequently used split-read aligners. AVAILABILITY: lack is distributed together with segemehl and freely available at www.bioinf.uni-leipzig.de/Software/segemehl/.

MOTIVATION: Tiling arrays have been a mainstay of unbiased genome-wide transcriptomics over the last decade. Currently available approaches to identify expressed or differentially expressed segments in tiling array data are limited in the recovery of the underlying gene structures and require several parameters that are intensity-related or partly dataset-specific. RESULTS: We have developed TileShuffle, a statistical approach that identifies transcribed and differentially expressed segments as significant differences from the background distribution while considering sequence-specific affinity biases and cross-hybridization. It avoids dataset-specific parameters in order to provide better comparability of different tiling array datasets, based on different technologies or array designs. TileShuffle detects highly and differentially expressed segments in biological data with significantly lower false discovery rates under equal sensitivities than commonly used methods. Also, it is clearly superior in the recovery of exon-intron structures. It further provides window z-scores as a normalized and robust measure for visual inspection. AVAILABILITY: The R package including documentation and examples is freely available at http://www.bioinf.uni-leipzig.de/Software/TileShuffle/

MOTIVATION: Cytosine DNA methylation is one of the major epigenetic modifications and influences gene expression, developmental processes, X-chromosome inactivation, and genomic imprinting. Aberrant methylation is furthermore known to be associated with several diseases including cancer. The gold standard to determine DNA methylation on genome-wide scales is 'bisulfite sequencing': DNA fragments are treated with sodium bisulfite resulting in the conversion of unmethylated cytosines into uracils, whereas methylated cytosines remain unchanged. The resulting sequencing reads thus exhibit asymmetric bisulfite-related mismatches and suffer from an effective reduction of the alphabet size in the unmethylated regions, rendering the mapping of bisulfite sequencing reads computationally much more demanding. As a consequence, currently available read mapping software often fails to achieve high sensitivity and in many cases requires unrealistic computational resources to cope with large real-life datasets. RESULTS: In this study, we present a seed-based approach based on enhanced suffix arrays in conjunction with Myers bit-vector algorithm to efficiently extend seeds to optimal semi-global alignments while allowing for bisulfite-related substitutions. It outperforms most current approaches in terms of sensitivity and performs time-competitive in mapping hundreds of millions of sequencing reads to vertebrate genomes. AVAILABILITY: The software segemehl is freely available at http://www.bioinf.uni-leipzig.de/Software/segemehl.

With few exceptions, current methods for short read mapping make use of simple seed heuristics to speed up the search. Most of the underlying matching models neglect the necessity to allow not only mismatches, but also insertions and deletions. Current evaluations indicate, however, that very different error models apply to the novel high-throughput sequencing methods. While the most frequent error-type in Illumina reads are mismatches, reads produced by 454's GS FLX predominantly contain insertions and deletions (indels). Even though 454 sequencers are able to produce longer reads, the method is frequently applied to small RNA (miRNA and siRNA) sequencing. Fast and accurate matching in particular of short reads with diverse errors is therefore a pressing practical problem. We introduce a matching model for short reads that can, besides mismatches, also cope with indels. It addresses different error models. For example, it can handle the problem of leading and trailing contaminations caused by primers and poly-A tails in transcriptomics or the length-dependent increase of error rates. In these contexts, it thus simplifies the tedious and error-prone trimming step. For efficient searches, our method utilizes index structures in the form of enhanced suffix arrays. In a comparison with current methods for short read mapping, the presented approach shows significantly increased performance not only for 454 reads, but also for Illumina reads. Our approach is implemented in the software segemehl available at http://www.bioinf.uni-leipzig.de/Software/segemehl/.