Generative Adversarial Networks (GAN) have achieved big success in various domains such as image generation, music generation, and natural language generation. In this paper, we propose a novel GAN-based collaborative filtering (CF) framework to provide higher accuracy in recommendation. We first identify a fundamental problem of existing GAN-based methods in CF and highlight it quantitatively via a series of experiments. Next, we suggest a new direction of vector-wise adversarial training to solve the problem and propose our GAN-based CF framework, called CFGAN, based on the direction. We identify a unique challenge that arises when vector-wise adversarial training is employed in CF. We then propose three CF methods realized on top of our CFGAN that are able to address the challenge. Finally, via extensive experiments on real-world datasets, we validate that vector-wise adversarial training employed in CFGAN is really effective to solve the problem of existing GAN-based CF methods. Furthermore, we demonstrate that our proposed CF methods on CFGAN provide recommendation accuracy consistently and universally higher than those of the state-of-the-art recommenders.

Existing review-aware recommendation methods represent users (or items) through the concatenation of the reviews written by (or for) them, and depend entirely on convolutional neural networks (CNNs) to extract meaningful features for modeling users (or items). However, understanding reviews based only on the raw words of reviews is challenging because of the inherent ambiguity contained in them originated from the users' different tendency in writing. Moreover, it is inefficient in time and memory to model users/items by the concatenation of their associated reviews owing to considerably large inputs to CNNs. In this work, we present a scalable review-aware recommendation method, called SentiRec, that is guided to incorporate the sentiments of reviews when modeling the users and the items. SentiRec is a two-step approach composed of the first step that includes the encoding of each review into a fixed-size review vector that is trained to embody the sentiment of the review, followed by the second step that generates recommendations based on the vector-encoded reviews. Through our experiments, we show that SentiRec not only outperforms the existing review-aware methods, but also drastically reduces the training time and the memory usage. We also conduct a qualitative evaluation on the vector-encoded reviews trained by SentiRec to demonstrate that the overall sentiments are indeed encoded therein.

Millions of posts are being generated in real-time by users in social networking services, such as Twitter. However, a considerable number of those posts are mundane posts that are of interest to the authors and possibly their friends only. This paper investigates the problem of automatically discovering valuable posts that may be of potential interest to a wider audience. Specifically, we model the structure of Twitter as a graph consisting of users and posts as nodes and retweet relations between the nodes as edges. We propose a variant of the HITS algorithm for producing a static ranking of posts. Experimental results on real world data demonstrate that our method can achieve better performance than several baseline methods.

Traditional ranking models for information retrieval lack the ability to make a clear distinction between relevant and nonrelevant documents at top ranks if both have similar bag-of-words representations with regard to a user query. We aim to go beyond the bag-of-words approach to document ranking in a new perspective, by representing each document as a sequence of sentences. We begin with an assumption that relevant documents are distinguishable from nonrelevant ones by sequential patterns of relevance degrees of sentences to a query. We introduce the notion of relevance flow, which refers to a stream of sentence-query relevance within a document. We then present a framework to learn a function for ranking documents effectively based on various features extracted from their relevance flows and leverage the output to enhance existing retrieval models. We validate the effectiveness of our approach by performing a number of retrieval experiments on three standard test collections, each comprising a different type of document: news articles, medical references, and blog posts. Experimental results demonstrate that the proposed approach can improve the retrieval performance at the top ranks significantly as compared with the state-of-the-art retrieval models regardless of document type.

Opinion retrieval involves the measuring of opinion score of a document about the given topic. We propose a new method, namely sentiment-relevance flow, that naturally unifies the topic relevance and the opinionated nature of a document. Experiments conducted over a large-scaled Web corpus show that the proposed approach improves performance of opinion retrieval in terms of precision at top ranks.

Most traditional ranking models roughly score the relevance of a given document by observing simple term statistics, such as the occurrence of query terms within the document or within the collection. Intuitively, the relative importance of query terms with regard to other individual non-query terms in a document can also be exploited to promote the ranks of documents in which the query is dedicated as the main topic. In this paper, we introduce a simple technique named intra-document term ranking, which involves ranking all the terms in a document according to their relative importance within that particular document. We demonstrate that the information regarding the rank positions of given query terms within the intra-document term ranking can be useful for enhancing the precision of top-retrieved results by traditional ranking models. Experiments are conducted on three standard TREC test collections.

With the rise of digital video consumptions, contextual video advertising demands have been increasing in recent years. This paper presents a novel video advertising system that selects relevant text ads for a given video scene by automatically identifying the situation of the scene. The situation information of video scenes is inferred from available video scripts. Experimental results show that the use of the situation information enhances the accuracy of ad retrieval for video scenes. The proposed system represents one of the pioneer video advertising systems using contextual information obtained from video scripts.

A key to success to contextual in-video advertising is finding advertising keywords on video contents effectively, but there has been little literature in the area so far. This paper presents some preliminary results of our learning-based system that finds relevant advertising keywords on particular scene of video contents using their scripts. The system is trained with not only features proven useful in earlier studies but novel features that reflect the situation of a targeted scene. Experimental results show that the new features are potentially helpful for enhancing the accuracy of keyword extraction for contextual in-video advertising.