Doreen Cheng

SeeNSearch: A Context Directed Search Facilitator for Home Entertainment Devices

The Internet has become an extremely popular source of entertainment and information. But, despite the growing amount of media content, most Web sites today are designed for access via web browsers on the PC, making it difficult for home consumers to access Internet content on their TVs or other devices that lack keyboards. As a result, the Internet is generally restricted to access on the PC or via cumbersome interfaces on non-PC devices. In this paper, we present unobtrusive and assistive technologies enabling home users to easily find and access Internet content related to the TV program they are watching. Using these technologies, the user is now able to access relevant information and video content on the Internet while watching TV.

Mobile Situation-Aware Task Recommendation Application

With more and more applications available on mobile devices, it has become increasingly difficult for users to find a desired application. Although research has been conducted for situation-awarere commendations on mobile devices, none addresses this problem; most research is for media content recommendations. Moreover, existing approaches assume predefined situations and/or user-specified profiles; some require users to intentionally train their devices before using them for recommendations. We believe that what defines a situation and what applications are preferred in the situation not only vary from user to user but also change over time, and therefore these assumptions and requirements are impractical for ordinary consumers. In this paper, we will describe our approach of using unsupervised learning, specifically co-clustering, to derive latent situation-based patterns from usage logs of user interactions with the device and environments and use the patterns for task and communication mode recommendations.

Clustering for Simultaneous Extraction of Aspects and Features from Reviews

This paper presents a clustering approach that simultaneously identifies product features and groups them into aspect categories from online reviews. Unlike prior approaches that first extract features and then group them into categories, the proposed approach combines feature and aspect discovery instead of chaining them. In addition, prior work on feature extraction tends to require seed terms and focus on identifying explicit features, while the proposed approach extracts both explicit and implicit features, and does not require seed terms. We evaluate this approach on reviews from three domains. The results show that it outperforms several state-of-the-art methods on both tasks across all three domains.

Nnon-collaborative interest mining for personal devices

In our daily life we frequently use mobile devices to interact with the people and things on the Internet. However, finding the right things when needed is getting difficult and frustrating. In this paper, we introduce a relatively new problem of non-collaborative personal interest mining using contexts and ratings available for items of interest. We present multi-step algorithms to extract personal situational interests from mobile phone usage logs without depending on other peoples data. The algorithms are based on clustering or a direct analogy from collaborative filtering. We provide extensive experimental results with our accuracy measure for synthetic data sets. The main advantages of our algorithms are: 1) no need for the user to train the phone actively, 2) no need for prior knowledge of the situations contained in a data set, 3) light-weight and running completely on a personal mobile phone and 4) good performance over low data densities. We also present a SmartSearch application. Upon user request, it automatically constructs search queries based on learned user interests and obtains information and advertisements for the user that suit the users situation.

Samsung: Align-and-Differentiate Approach to Semantic Textual Similarity

This paper describes our Align-andDifferentiate approach to the SemEval 2015 Task 2 competition for English Semantic Textual Similarity (STS) systems. Our submission achieved the top place on two of the five evaluation datasets. Our team placed 3rd among 28 participating teams, and our three runs ranked 4th, 6th and 7th among the 73 runs submitted by the 28 teams. Our approach improves upon the UMBC PairingWords system by semantically differentiating distributionally similar terms. This novel addition improves results by 2.5 points on the Pearson correlation measure.

Active Learning with Efficient Feature Weighting Methods for Improving Data Quality and Classification Accuracy

Many machine learning datasets are noisy with a substantial number of mislabeled instances. This noise yields sub-optimal classification performance. In this paper we study a large, low quality annotated dataset, created quickly and cheaply using Amazon Mechanical Turk to crowdsource annotations. We describe computationally cheap feature weighting techniques and a novel non-linear distribution spreading algorithm that can be used to iteratively and interactively correcting mislabeled instances to significantly improve annotation quality at low cost. Eight different emotion extraction experiments on Twitter data demonstrate that our approach is just as effective as more computationally expensive techniques. Our techniques save a considerable amount of time.

SmartSearch: Situation-Aware Web Search on Mobile Devices

To provide consumers with the right information at the time of need, we developed a SmartSearch application that is able to extract a users situational interests from usage data. It automatically constructs search queries based on situational interests. We extract the situational interests automatically without prior training and user involvement. In addition, SmartSearch is a client-side solution running completely on a mobile device to protect users privacy.

TISA: topic independence scoring algorithm

Textual analysis using machine learning is in high demand for a wide range of applications including recommender systems, business intelligence tools, and electronic personal assistants. Some of these applications need to operate over a wide and unpredictable array of topic areas, but current in-domain, domain adaptation, and multi-domain approaches cannot adequately support this need, due to their low accuracy on topic areas that they are not trained for, slow adaptation speed, or high implementation and maintenance costs. To create a true domain-independent solution, we introduce the Topic Independence Scoring Algorithm (TISA) and demonstrate how to build a domain-independent bag-of-words model for sentiment analysis. This model is the best preforming sentiment model published on the popular 25 category Amazon product reviews dataset. The model is on average 89.6% accurate as measured on 20 held-out test topic areas. This compares very favorably with the 82.28% average accuracy of the 20 baseline in-domain models. Moreover, the TISA model is highly uniformly accurate, with a variance of 5 percentage points, which provides strong assurance that the model will be just as accurate on new topic areas. Consequently, TISAs models are truly domain independent. In other words, they require no changes or human intervention to accurately classify documents in never before seen topic areas.

Theme issue on adaptation and personalization for ubiquitous computing

Ubiquitous computing is a human-centered paradigm that aims to provide users with adaptive and personalized services according to their surrounding context. Adaptation and personalization technologies are thus an important basis of ubiquitous computing. They are also the core for realizing context awareness in pervasive service provisioning. In ubiquitous computing environments, people are surrounded by many networked computers, both fixed (e.g., PCs, TVs) and mobile devices such as PDAs, cellular phones, etc. People are increasingly able to access their desired content, anytime, anywhere using the available devices. To offer the right information to users at the right time, right place and in the right way is challenging for many reasons, such as varying user interests, heterogeneous environments and devices, dynamic networks, information overload, user privacy, and so on. This theme issue aims to explore adaptation and personalization services and technologies for ubiquitous computing. Submissions to this special issue came from an open call for papers as well as from selected papers presented at the 7th International Conference on Ubiquitous Intelligence and Computing (UIC 2010) held at Xi’an, China, October 26–29, 2010. We received a total of 26 submissions of which 8 papers were accepted after three rounds of rigorous reviews. We are grateful to the large number of reviewers who assisted us in the review process; in order to ensure high reviewing standards, three to four reviewers evaluated each paper. The opening paper of this special issue, ‘‘Social itinerary recommendation from user-generated digital trails’’, authored by Hyoseok Yoon, Yu Zheng, Xing Xie, and Woontack Woo received the best paper award of UIC 2010. The paper addresses the problem of planning travel to unfamiliar regions for novice travelers. It proposes recommending a social itinerary by learning from multiple user-generated digital trails, such as GPS trajectories of residents and travel experts. It describes an itinerary model in terms of attributes extracted from user-generated GPS trajectories, and a social itinerary recommendation framework that can find and rank itinerary candidates. It also reports the evaluation results using a large set of usergenerated GPS trajectories collected from Beijing, China. The second paper, ‘‘TruBeRepec: a trust-behavior-based reputation and recommender system for mobile applications’’, by Zheng Yan, Peng Zhang, and Robert H. Deng, examines the trustworthiness of mobile applications that are to be recommended to a user. The authors introduce a model of trust behavior for mobile applications based on the results of a large-scale user survey. Several algorithms Z. Yu (&) School of Computer Science, Northwestern Polytechnical University, Xian, China e-mail: zhiwenyu@nwpu.edu.cn

Situation-Aware on mobile phone using co-clustering: algorithms and extensions

Due to the large number of applications in the mobile phones, users usually go through a fixed menu hierarchy to find a specific interesting application. Hence, in our previous research, we realized the proactive mobile phone application recommendation using co-clustering and demonstrated the promising recommendation performance on a smartphone. The approach first autonomously extracts users behavioral patterns from the usage log of user interactions with the device as well as environments and then recommends potential applications that might be interesting to the user at the corresponding specific situation. In this paper, as a follow-up to this novel platform of intelligent smartphone-based situation-awareness, we investigate sophisticated methodologies that lead to better performance. To achieve this goal, we considered various co-clustering algorithms with different data transformations and weighting schemes for simulated mobile phone usage data. Through non-exhaustive, but pretty comprehensive experimental setting, we find what specific co-clustering algorithms with what specific data transformations and weighting schemes improve accuracy performance in extracting specific user patterns.