Sequential recommendation is the task of predicting the next interacted item of a target user, given his/her past interaction sequence. Conventionally, sequential recommenders are evaluated offline with the last item in each sequence as the sole correct (relevant) label for the testing example of the corresponding user. However, little is known about how this sparsity of preference data affects the robustness of the offline evaluation's outcomes. To help researchers address this, we collect additional preference data via crowdsourcing. Specifically, we propose an assessment interface tailored to the sequential recommendation task and ask crowd workers to assess the (potential) relevance of each candidate item in MovieLens 1M, a commonly used dataset. Toward establishing a more robust evaluation methodology, we release the collected preference data, which we call ML-1M++, as well as the code of the assessment interface.

Large events with many attendees cause congestion in the traffic network around the venue. To avoid accidents or delays due to this kind of unexpected congestion, it is important to predict the level of congestion in advance of the event. This study aimed to forecast congestion triggered by large events. However, historical congestion information alone is insufficient to forecast congestion at large venues when non-recurrent events are held there. To address this problem, we utilize microblog posts that refer to future events as an indicator of event attendance. We propose a regression model that is trained with microblog posts and historical congestion information to accurately forecast congestion at large venues. Experiments on next 24-hour congestion forecasting using real-world traffic and Twitter data demonstrate that our model reduces the prediction errors over those of the baseline models (autoregressive and long short term memory) by 20% - 50%.

How can recommender systems help people improve their skills? As a first step toward recommendation for the upskilling of users, this paper addresses the problems of modeling the improvement of user skills and the difficulty of items in action sequences where users select items at different times. We propose a progression model that uses latent variables to learn the monotonically non-decreasing progression of user skills. Once this model is trained with the given sequence data, we leverage it to find a statistical solution to the item difficulty estimation problem, where we assume that users usually select items within their skill capacity. Experiments on five datasets (four from real domains, and one generated synthetically) revealed that (1) our model successfully captured the progression of domain-dependent skills; (2) multi-faceted item features helped to learn better models that aligned well with the ground-truth skill and difficulty levels in the synthetic dataset; (3) the learned models were practically useful to predict items and ratings in action sequences; and (4) exploiting the dependency structure of our skill model for parallel computation made the training process more efficient.

Understanding the spread of diseases and the use of medicines is of practical importance for various organizations, such as medical providers, medical payers, and national governments. This study aims to detect the change in the prescription trends and to identify its cause through an analysis of Medical Insurance Claims (MICs), which comprise the specifications of medical fees charged to health insurers. Our approach is two-fold. (1) We propose a latent variable model that simulates the medication behavior of physicians to accurately reproduce monthly prescription time series from the MIC data, where prescription links between the diseases and medicines are missing. (2) We apply a state space model with intervention variables to decompose the monthly prescription time series into different components including seasonality and structural changes. Using a large dataset consisting of 3.5-year MIC records, we conduct experiments to evaluate our approach in terms of accuracy, usefulness, and efficiency. We also demonstrate three applications for our medical analysis.

To address the difficulty in clipping articles from various mobile applications (apps), we propose a novel framework called UniClip, which allows a user to snap a screen of an article to save the whole article in one place. The key task of the framework is search by screenshots , which has three challenges: (1) how to represent a screenshot; (2) how to formulate queries for effective article retrieval; and (3) how to identify the article from search results. We solve these by (1) segmenting a screenshot into structural units called blocks, (2) formulating effective search queries by considering the role of each block, and (3) aggregating the search result lists of multiple queries. To improve efficiency, we also extend our approach with learning-to-rank techniques so that we can find the desired article with only one query. Experimental results show that our approach achieves high retrieval performance ( F 1 = 0.868), which outperforms baselines based on keyword extraction and chunking methods. Learning-to-rank models improve our approach without learning by about 6%. A user study conducted to investigate the usability of UniClip reveals that ours is preferred by 21 out of 22 participants for its simplicity and effectiveness.

For intrinsically diverse tasks, in which collecting extensive information from different aspects of a topic is required, searchers often have difficulty formulating queries to explore diverse aspects and deciding when to stop searching. With the goal of helping searchers discover unexplored aspects and find the appropriate timing for search stopping in intrinsically diverse tasks, we propose ScentBar, a query suggestion interface visualizing the amount of important information that a user potentially misses collecting from the search results of individual queries. We define the amount of missed information for a query as the additional gain that can be obtained from unclicked search results of the query, where gain is formalized as a set-wise metric based on aspect importance, aspect novelty, and per-aspect document relevance and is estimated by using a state-of-the-art algorithm for subtopic mining and search result diversification. Results of a user study involving 24 participants showed that the proposed interface had the following advantages when the gain estimation algorithm worked reasonably: (1) ScentBar users stopped examining search results after collecting a greater amount of relevant information; (2) they issued queries whose search results contained more missed information; (3) they obtained higher gain, particularly at the late stage of their sessions; and (4) they obtained higher gain per unit time. These results suggest that the simple query visualization helps make the search process of intrinsically diverse tasks more efficient, unless inaccurate estimates of missed information are visualized.

In this paper we address the difficulty of clipping articles from mobile apps. We propose a service called UniClip that allows a user to save the full content of an article by snapping a screenshot part of it. UniClip leverages a huge amount of indexed web data to mine the article by starting with a snapped screenshot. We propose approaches to solve three challenges: (1) how to represent a screenshot; (2) how to formulate effective queries for retrieving a full article; and (3) how to rank the best URL at the top from multiple search result lists. Experimental results indicate that our approach is effective in achieving as high an $$F_1$$ F 1 measure as 0.905, which outperforms the best of three baseline methods by 18 points.

In this paper, we propose a two-stage system using user's eye movements to accommodate the increasing demands to obtain information from the Web in an efficient way. In the first stage the system estimates a user's search intent as a set of weighted terms extracted based on the user's eye movements while browsing Web pages. Then in the second stage, the system shows relevant information to the user by using the estimated intent for re-ranking search results, suggesting intent-based queries, and emphasizing relevant parts of Web pages. The system aims to help users to efficiently obtain what they need by repeating these steps throughout the information seeking process. We proposed four types of search intent estimation methods (MLT, nMLT, DLT and nDLT) considering the relationship among intents, term frequencies and eye movements. As a result of an experiment designed for evaluating the accuracy of each method with a prototype system, we confirmed that the nMLT method works best. In addition, by analyzing the extracted intent terms for eight subjects in the experiment, we found that the system could estimate the unique search intent of each user even if they performed the same search tasks.