Knowledge graph (KG) is the backbone to enable cognitive Artificial Intelligence (AI), which relies on cognitive computing and semantic reasoning. Knowledge graph is the connected data with the semantically enriched context. It is the necessary step for the next move of AI. Our daily activities have closely intermingled with various applications powered by knowledge graphs. It has even entered our healthcare system to facilitate clinical decision making and improve hospital efficiency. This workshop aims to bring researchers and practitioners to promote research and applications related to knowledge graph.

Food recommendation has become an important means to help guide users to adopt healthy dietary habits. Previous works on food recommendation either i) fail to consider users' explicit requirements, ii) ignore crucial health factors (e.g., allergies and nutrition needs), or iii) do not utilize the rich food knowledge for recommending healthy recipes. To address these limitations, we propose a novel problem formulation for food recommendation, modeling this task as constrained question answering over a large-scale food knowledge base/graph (KBQA). Besides the requirements from the user query, personalized requirements from the user's dietary preferences and health guidelines are handled in a unified way as additional constraints to the QA system. To validate this idea, we create a QA style dataset for personalized food recommendation based on a large-scale food knowledge graph and health guidelines. Furthermore, we propose a KBQA-based personalized food recommendation framework which is equipped with novel techniques for handling negations and numerical comparisons in the queries. Experimental results on the benchmark show that our approach significantly outperforms non-personalized counterparts (average 59.7% absolute improvement across various evaluation metrics), and is able to recommend more relevant and healthier recipes.

Although it has become easier for individuals to track their personal health data (e.g., heart rate, step count, and nutrient intake data), there is still a wide chasm between the collection of data and the generation of meaningful summaries to help users better understand what their data means to them. With an increased comprehension of their data, users will be able to act upon the newfound information and work toward striving closer to their health goals. We aim to bridge the gap between data collection and summary generation by mining the data for interesting behavioral findings that may provide hints about a user’s tendencies. Our focus is on improving the explainability of temporal personal health data via a set of informative summary templates, or “protoforms.” These protoforms span both evaluation-based summaries that help users evaluate their health goals and pattern-based summaries that explain their implicit behaviors. In addition to individual-level summaries, the protoforms we use are also designed for population-level summaries. We apply our approach to generate summaries (both univariate and multivariate) from real user health data and show that the summaries our system generates are both interesting and useful.

In recent years, there has been growing interest in the use of fitness trackers and smartphone applications for promoting physical activity. Many of these applications use accelerometers to estimate the level of activity that users engage in and provide visual reports of a user's step counts. When provided, most recommendations are limited to popular general health advice. In our study, we develop an approach for providing data-driven and personalized recommendations for intraday activity planning. We generate an hour-by-hour activity plan that is based on the user's probability of adhering to the plan. The user's probability of adherence to the plan is personalized, based on his/her past activity patterns and current activity target. Using this approach, we can tailor notifications (e.g., reminders, encouragement) to each user. We can also dynamically update the user's activity plan at mid-day, if his/her actual activity deviates sufficiently from the original plan. In this paper, we describe an implementation of our approach and report our technical findings with respect to identifying typical activity patterns from historical data, predicting whether an activity target will be achieved, and adapting an activity plan based on a user's actual performance throughout the day.

Multiple description coding (MDC) decomposes one single media into several descriptions and transmits them over different channels for error resilience. Each description contributes to improving the reconstructed media quality when decoded. Distributed video coding (DVC) encodes multiple correlated images and utilizes error correction codes to shift the codec complexity to a joint decoder. Combining MDC with DVC (MDVC) yields a stable codec for mobile encoders. In this paper, to improve the MDVC codec performance, image correlations among the MDVC processing modules were exploited to improve reconstructed video quality and enhance transmission robustness. At the side encoder, a DVC-based adaptive differential pulse code modulation was designed to remove interframe redundancy to enhance rate-distortion performances. For the MDVC central decoding, intradescription and interdescription correlations were utilized to dynamically select the best reconstructed frames from two descriptions, instead of selecting just one description or all key-frames from two descriptions. Experiments showed that, as compared to previous methods, the proposed MDVC control method yielded 1-2 dB higher in image PSNRs for Wyner-Ziv reconstructed frames at the side decoder when encoding low-to-medium complexity videos. For high-complexity videos, it effectively prevents error correction of Wyner-Ziv frames from malfunctioning and yields about 3 dB higher in PSNR. The proposed MDVC central decoder control yields 1-4 dB higher PSNRs, as compared to side decoders. Under lossy transmission, it demonstrates 27-64% smaller PSNR variations, as compared to that of combining key-frames as the decoded video. The proposed MDVC system and control not only improve the DVC reconstructed video quality, but also reduce the quality fluctuation artifacts of MDC coded video for mobile coders.

With the advance of multi-view video codec (MVC) technology, the multimedia platform can display videos of different views. The distributed video coder (DVC) that adopts the Wynzer-Ziv (WZ) codec can shift encoding complexity to decoder under the MVC framework, denoted as multi-view DVC (MDVC), for possible low-complexity encoder applications. For side information (SI) generation, the intra- and inter-view video correlations among images in MDVC were exploited to improve the SI confidence to yield better reconstructed WZ images. We proposed a new block matching and prediction (BMP) algorithm based on Scale Invariant Feature Transform, abbreviated as SIFT-BMP, to yield more accurate SI to improve the reconstructed video quality. Simulations verified that the proposed SIFT-BMP yielded more accurate SI and higher PSNR of reconstructed images, as compared to previous methods. The PSNR improvements are 1.6~2.5 dB and higher, as compared to MCTI and H.264/AVC, for medium to high complexity videos.

Multiple description coding (MDC) provides stable wireless multimedia communications with the help of multiple transmission paths, while distributed video coding (DVC) effectively shifts the encoder complexity to decoder for mobile media devices. We proposed to combine MDC with DVC, abbreviated as MDVC, to construct a reliable and efficient video codec system. To improve the codec performance, the DPCM prediction gain is exploited from the correlations among sub-video sequences inside the MDVC to increase the reconstructed video quality. In addition, a MDVC central decoder control method, operated dynamically at frame-level, is proposed to improve the adaptation capability, as compared to the current MDC central decoder that selects one best description for the output video. The intra-and inter-correlations among MDC descriptions were investigated to make this frame selection decision for the central decoder. Simulations showed that the proposed MDVC control method demonstrates higher average PSNR and lower PSNR variations of the output videos, with the help of DPCM prediction gain and the efficient central decoder control method that correctly selected high quality WZ frames for low quality key-frames.