This study explores the problem of inferring locations for individual tweets. We scrutinize Twitter user timelines in a novel fashion. First of all, we split each user's tweet timeline temporally into a number of clusters, each tending to imply a distinct location. Subsequently, we adapt machine learning models to our setting and design classifiers that classify each tweet cluster into one of the pre-defined location classes at the city level. Extensive experiments on a large set of real Twitter data suggest that our models are effective at inferring locations for non-geotagged tweets and outperform the state-of-the-art approaches significantly in terms of inference accuracy.

Social media like Twitter have become globally popular in the past decade. Thanks to the high penetration of smartphones, social media users are increasingly going mobile. This trend has contributed to foster various location based services deployed on social media, the success of which heavily depends on the availability and accuracy of users' location information. However, only a very small fraction of tweets in Twitter are geo-tagged. Therefore, it is necessary to infer locations for tweets in order to attain the purpose of those location based services. In this paper, we tackle this problem by scrutinizing Twitter user timelines in a novel fashion. First of all, we split each user's tweet timeline temporally into a number of clusters, each tending to imply a distinct location. Subsequently, we adapt two machine learning models to our setting and design classifiers that classify each tweet cluster into one of the pre-defined location classes at the city level. The Bayes based model focuses on the information gain of words with location implications in the user-generated contents. The convolutional LSTM model treats user-generated contents and their associated locations as sequences and employs bidirectional LSTM and convolution operation to make location inferences. The two models are evaluated on a large set of real Twitter data. The experimental results suggest that our models are effective at inferring locations for non-geotagged tweets and the models outperform the state-of-the-art and alternative approaches significantly in terms of inference accuracy.

Transfer learning aims at adapting a model learned from source dataset to target dataset. It is a beneficial approach especially when annotating on the target dataset is expensive or infeasible. Transfer learning has demonstrated its powerful learning capabilities in various vision tasks. Despite transfer learning being a promising approach, it is still an open question how to adapt the model learned from the source dataset to the target dataset. One big challenge is to prevent the impact of category bias on classification performance. Dataset bias exists when two images from the same category, but from different datasets, are not classified as the same. To address this problem, a transfer learning algorithm has been proposed, called negative back‐dropout transfer learning (NB‐TL), which utilizes images that have been misclassified and further performs back‐dropout strategy on them to penalize errors. Experimental results demonstrate the effectiveness of the proposed algorithm. In particular, the authors evaluate the performance of the proposed NB‐TL algorithm on UCF 101 action recognition dataset, achieving 88.9% recognition rate.

Wikipedia is one of the fastest growing websites and a primary source of knowledge on the Internet. Being a wiki, its content is crowd-sourced by the users. This has many benefits and it is one of the main reasons it has grown to reach more than 5 million articles in its English version. Nevertheless, this also raises issues, like the overlinking of articles, which are difficult to deal with by editors. In this paper, we tackle overlinking in Wikipedia as a ranking problem. We apply Learning to Rank algorithms to evaluate the click frequency of links in an effort to distinguish the most useful links for users. To accomplish this, we develop a ground truth, which serves as baseline for our algorithm and compare hyperlink features to implement the most advantageous ones. The results show 86.2% accuracy with the top-6 most useful features and 87.7% accuracy with the complete feature set. Considering these results, we outline a solution to the overlinking problem. By removing the most inadequate links, we suggest that readability of Wikipedia articles could be improved while preserving most of its useful links.

With the growing amount of content stored in personal and organizational information spaces, finding and re-finding documents becomes both more crucial and challenging. In this work, we propose an approach to reduce information overload in navigation by automatically focusing on important documents, adaptively to the tasks at hand. Based on the idea of managed forgetting, we present a ranking method, which unifies activity logs and semantic information about documents into a common framework to identify important documents to the user's current tasks. Our experiments on two real-world datasets, both collected from knowledge work activities in professional scenarios, show that our ranking approach outperforms the baseline methods for both subsequent access prediction and the effectiveness in ranking important documents. Furthermore, we implemented and demonstrated a system for decluttering information spaces as a proof of concept of our managed forgetting approach.

Twitter has become a valuable source of event-related information, namely, breaking news and local event reports. Due to its capability of transmitting information in real-time, Twitter is further exploited for timeline summarisation of high-impact events, such as protests, accidents, natural disasters or disease outbreaks. Such summaries can serve as important event digests where users urgently need information, especially if they are directly affected by the events. In this paper, we study the problem of timeline summarisation of high-impact events that need to be generated in real-time. Our proposed approach includes four stages: classification of real-world events reporting tweets, online incremental clustering, post-processing and sub-events summarisation. We conduct a comprehensive evaluation of different stages on the “Ebola outbreak” tweet stream, and compare our approach with several baselines, to demonstrate its effectiveness. Our approach can be applied as a replacement of a manually generated timeline and provides early alarms for disaster surveillance.

The study of temporal dynamics and its impact can be framed within the so-called temporal IR approaches, which explain how user behavior, document content and scale vary with time, and how we can use them in our favor in order to improve retrieval effectiveness.

Long-running, high-impact events such as the Boston Marathon bombing often develop through many stages and involve a large number of entities in their unfolding. Timeline summarization of an event by key sentences eases story digestion, but does not distinguish between what a user remembers and what she might want to re-check. In this work, we present a novel approach for timeline summarization of high-impact events, which uses entities instead of sentences for summarizing the event at each individual point in time. Such entity summaries can serve as both (1) important memory cues in a retrospective event consideration and (2) pointers for personalized event exploration. In order to automatically create such summaries, it is crucial to identify the "right" entities for inclusion. We propose to learn a ranking function for entities, with a dynamically adapted trade-off between the in-document salience of entities and the informativeness of entities across documents, i.e., the level of new information associated with an entity for a time point under consideration. Furthermore, for capturing collective attention for an entity we use an innovative soft labeling approach based on Wikipedia. Our experiments on a real large news datasets confirm the effectiveness of the proposed methods.

When selecting important photos from a personal photo collection - e.g. for creating an enjoyable sub-collection for revisiting or preservation - photos are not considered in isolation. Therefore, collection-level criteria are also taken into account by automated photo selection methods. However, the typical two-step process of first clustering and subsequently picking from the clusters seems to overstress coverage as a criterion when applied to the task of selecting the photos most important to a user. We, therefore, propose a novel expectation-oriented photo selection method, which combines a variety of collection-level and image-level selection criteria in a flexible way. In our evaluation, which is based on large real-world personal photo collections with overall more than 18,000 images, we show that our method outperforms state-of-the-art photo selection methods. In addition, the proposed method does not rely on any manual annotations, making it applicable in realistic settings of personal photo collections.

Due to their first-hand, diverse and evolution-aware reflection of nearly all areas of life, web archives are emerging as gold-mines for content analytics of many sorts. However, supporting search, which goes beyond navigational search via URLs, is a very challenging task in these unique structures with huge, redundant and noisy temporal content. In this paper, we address the search needs of expert users such as journalists, economists or historians for discovering a topic in time: Given a query, the top-k returned results should give the best representative documents that cover most interesting time-periods for the topic. For this purpose, we propose a novel random walk-based model that integrates relevance, temporal authority, diversity and time in a unified framework. Our preliminary experimental results on the large-scale real-world web archival collection shows that our method significantly improves the state-of-the-art algorithms (i.e., PageRank) in ranking temporal web pages.

Fully understanding an older news article requires context knowledge from the time of article creation. Finding information about such context is a tedious and time-consuming task, which distracts the reader. Simple contextualization via Wikification is not sufficient here. The retrieved context information has to be time-aware, concise (not full Wikipages) and focused on the coherence of the article topic. In this paper, we present an approach for time-aware recontextualization, which takes those requirements into account in order to improve reading experience. For this purpose, we propose (1) different query formulation methods for retrieving contextualization candidates and (2) ranking methods taking into account topical and temporal relevance as well as complementarity with respect to the original text. We evaluate our proposed approaches through extensive experiments using real-world datasets and ground-truth consisting of over 9,400 article/context pairs. To this end, our experimental results show that our approaches retrieve contextualization information for older articles from the New York Times Archive with high precision and outperform baselines significantly.

Understanding a text, which was written some time ago, can be compared to translating a text from another language. Complete interpretation requires a mapping, in this case, a kind of time-travel translation between present context knowledge and context knowledge at time of text creation. In this paper, we study time-aware re-contextualization, the challenging problem of retrieving concise and complementing information in order to bridge this temporal context gap. We propose an approach based on learning to rank techniques using sentence-level context information extracted from Wikipedia. The employed ranking combines relevance, complimentarity and time-awareness. The effectiveness of the approach is evaluated by contextualizing articles from a news archive collection using more than 7,000 manually judged relevance pairs. To this end, we show that our approach is able to retrieve a significant number of relevant context information for a given news article.

A large number of mainstream applications, like temporal search, event detection, and trend identification, assume knowledge of the timestamp of every document in a given textual collection. In many cases, however, the required timestamps are either unavailable or ambiguous. A charac- teristic instance of this problem emerges in the context of large repositories of old digitized documents. For such doc- uments, the timestamp may be corrupted during the digiti- zation process, or may simply be unavailable. In this paper, we study the task of approximating the timestamp of a doc- ument, so-called document dating. We propose a content- based method and use recent advances in the domain of term burstiness, which allow it to overcome the drawbacks of pre- vious document dating methods, e.g. the fix time partition strategy. We use an extensive experimental evaluation on different datasets to validate the efficacy and advantages of our methodology, showing that our method outperforms the state of the art methods on document dating.

Retrieval effectiveness of temporal queries can be improved by taking into account the time dimension. Existing temporal ranking models follow one of two main approaches: 1) a mixture model linearly combining textual similarity and temporal similarity, and 2) a probabilistic model generating a query from the textual and temporal part of document independently. In this paper, we propose a novel time-aware ranking model based on learning-to-rank techniques. We employ two classes of features for learning a ranking model, entity-based and temporal features, which are derived from annotation data. Entity-based features are aimed at capturing the semantic similarity between a query and a document, whereas temporal features measure the temporal similarity. Through extensive experiments we show that our ranking model significantly improves the retrieval effectiveness over existing time-aware ranking models.

News prediction retrieval has recently emerged as the task of retrieving predictions related to a given news story (or a query). Predictions are defined as sentences containing time references to future events. Such future-related information is crucially important for understanding the temporal development of news stories, as well as strategies planning and risk management. The aforementioned work has been shown to retrieve a significant number of relevant predictions. However, only a certain news topics achieve good retrieval effectiveness. In this paper, we study how to determine the difficulty in retrieving predictions for a given news story. More precisely, we address the query difficulty estimation problem for news prediction retrieval. We propose different entity-based predictors used for classifying queries into two classes, namely, Easy and Difficult. Our prediction model is based on a machine learning approach. Through experiments on real-world data, we show that our proposed approach can predict query difficulty with high accuracy.

Microblogging services, such as Twitter, are gaining interests as a means of sharing information in social networks. Numerous works have shown the potential of using Twitter posts (or tweets) in order to infer the existence and magnitude of real-world events. In the medical domain, there has been a surge in detecting public health related tweets for early warning so that a rapid response from health authorities can take place. In this paper, we present a temporal analytics tool for supporting a comparative, temporal analysis of disease outbreaks between Twitter and official sources, such as, World Health Organization (WHO) and ProMED-mail. We automatically extract and aggregate outbreak events from official outbreak reports, producing time series data. Our tool can support a correlation analysis and an understanding of the temporal developments of outbreak mentions in Twitter, based on comparisons with official sources.

Time-aware retrieval models exploit one of two time dimensions, namely, (a) publication time or (b) content time (temporal expressions mentioned in documents). We show that the effectiveness for a temporal query (e.g., illinois earthquake 1968) depends significantly on which time dimension is factored into ranking results. Motivated by this, we propose a machine learning approach to select the most suitable time-aware retrieval model for a given temporal query. Our method uses three classes of features obtained from analyzing distributions over two time dimensions, a distribution over terms, and retrieval scores within top-k result documents. Experiments on real-world data with crowdsourced relevance assessments show the potential of our approach.

We estimate that nearly one third of news articles contain references to future events. While this information can prove crucial to understanding news stories and how events will develop for a given topic, there is currently no easy way to access this information. We propose a new task to address the problem of retrieving and ranking sentences that contain mentions to future events, which we call ranking related news predictions. In this paper, we formally define this task and propose a learning to rank approach based on 4 classes of features: term similarity, entity-based similarity, topic similarity, and temporal similarity. Through extensive evaluations using a corpus consisting of 1.8 millions news articles and 6,000 manually judged relevance pairs, we show that our approach is able to retrieve a significant number of relevant predictions related to a given topic.

Query performance prediction is aimed at predicting the retrieval effectiveness that a query will achieve with respect to a particular ranking model. In this paper, we study query performance prediction for a ranking model that explicitly incorporates the time dimension into ranking. Different time-based predictors are proposed as analogous to existing keyword-based predictors. In order to improve predicting performance, we combine different predictors using linear regression and neural networks. Extensive experiments are conducted using queries and relevance judgments obtained by crowdsourcing.

When searching a temporal document collection, e.g., news archives or blogs, the time dimension must be explicitly incorporated into a retrieval model in order to improve relevance ranking. Previous work has followed one of two main approaches: 1) a mixture model linearly combining textual similarity and temporal similarity, or 2) a probabilistic model generating a query from the textual and temporal part of a document independently. In this paper, we compare the effectiveness of different time-aware ranking methods by using a mixture model applied to all methods. Extensive evaluation is conducted using the New York Times Annotated Corpus, queries and relevance judgments obtained using the Amazon Mechanical Turk.

In a text retrieval community, many researchers have shown a good quality of searching a current snapshot of the Web. However, only a small number have demonstrated a good quality of searching a long-term archival domain, where documents are preserved for a long time, i.e., ten years or more. In such a domain, a search application is not only applicable for archivists or historians, but also in a context of national library and enterprise search (searching document repositories, emails, etc.). In the rest of this paper, we will explain three problems of searching document archives and propose possible approaches to solve these problems. Our main research question is: How to improve the quality of search in a document archive using temporal information?