Multimodal sentiment analysis for fashion-related social media is essential for understanding how consumers appraise fashion products across platforms like Instagram and Twitter, where both textual and visual elements contribute to sentiment expression. However, a notable challenge in this task is the modality gap, where the different information density between text and images hinders effective sentiment analysis. In this paper, we propose a novel multimodal framework that addresses this challenge by introducing pseudo data generated by a two-stage framework. We further utilize a multimodal fusion approach that efficiently integrates the information from various modalities for sentiment classification of fashion posts. Experiments conducted on a comprehensive dataset demonstrate that our framework significantly outperforms existing unimodal and multimodal baselines, highlighting its effectiveness in bridging the modality gap for more accurate sentiment classification in fashion-related social media posts.

The related work section is important in academic papers, synthesizing current research progress and contextualizing new studies. However, it is challenging to automatically generate such a section due to the complexity of citation networks and domain-specific differences. Addressing this, we introduce a novel method that employs contrastive learning to utilize the rich domain information from citation networks. Specifically, we design a Domain-Aware Related Work Generator (DAG) to aggregate domain information based on an advanced attention mechanism. This mechanism is essential for the effective combination of the related domain knowledge to the current research, ensuring that the related work generation is not only based on direct citations but also rich in contextual domain insights. Based on this mechanism, we can effectively identify and aggregate essential domain information related to the research topic. We construct a comprehensive dataset comprising various related work sections and their corresponding citation networks to validate our methodology. Our experimental results demonstrate a notable improvement over existing state-of-the-art methods, emphasizing the effectiveness of our model in generating more accurate and contextually enriched related work sections. These findings represent a significant advance in the automated generation of academically related work.

Question generation (QG) aims to create a fluent question from a passage and a target answer. State-of-the-art approaches are mainly based on encoder-decoder models to generate questions from the given passage and answer, which focus on using the information contained in a particular part of the passage for QG, but unaware of the clues hidden in other parts of the passage. Besides, the existing work on QG mainly focus on generating factoid questions, which are less suitable for generating non-factoid questions such as why-questions. In this paper, we propose to augment encoder-decoder framework with a pair-wise self-matching attention mechanism to dynamically collect inter-sentential evidence from the whole passage according to the current passage word and answer information. Besides, to let the model be more suitable for why-question generation, we also involve some causal features in the encoding process. Finally, to tackle the lack of why-question generation training data problem, we adopt a distant supervised method with an initial causal knowledge base to generate a large training data for why-question generation. Extensive experiments on several data sets show that our model significantly outperforms state-of-the-art question generation models not only on why-question generation tasks, but also on other types of question generation tasks.

Previous studies on temporal relation extraction focus on mining sentence-level information or enforcing coherence on different temporal relation types among various event mentions in the same sentence or neighboring sentences, largely ignoring those discourse-level temporal relations in nonadjacent sentences. In this paper, we propose a discourse-level global inference model to mine those temporal relations between event mentions in document-level, especially in nonadjacent sentences. Moreover, we provide various kinds of discourse-level constraints, which derived from event semantics, to further improve our global inference model. Evaluation on a Chinese corpus justifies the effectiveness of our discourse-level global inference model over two strong baselines.

Textual information is of critical importance for automatic user classification in social media. However, most previous studies model textual features in a single perspective while the text in a user homepage typically possesses different styles of text, such as original message and comment from others. In this paper, we propose a novel approach, namely ensemble LSTM, to user classification by incorporating multiple textual perspectives. Specifically, our approach first learns a LSTM representation with a LSTM recurrent neural network and then presents a joint learning method to integrating all naturally-divided textual perspectives. Empirical studies on two basic user classification tasks, i.e., gender classification and age classification, demonstrate the effectiveness of the proposed approach to user classification with multiple textual perspectives.

This article presents a unified framework for extracting standard and update summaries from a set of documents. In particular, a topic modeling approach is employed for salience determination and a dynamic modeling approach is proposed for redundancy control. In the topic modeling approach for salience determination, we represent various kinds of text units, such as word, sentence, document, documents, and summary, using a single vector space model via their corresponding probability distributions over the inherent topics of given documents or a related corpus. Therefore, we are able to calculate the similarity between any two text units via their topic probability distributions. In the dynamic modeling approach for redundancy control, we consider the similarity between the summary and the given documents, and the similarity between the sentence and the summary, besides the similarity between the sentence and the given documents, for standard summarization while for update summarization, we also consider the similarity between the sentence and the history documents or summary. Evaluation on TAC 2008 and 2009 in English language shows encouraging results, especially the dynamic modeling approach in removing the redundancy in the given documents. Finally, we extend the framework to Chinese multi-document summarization and experiments show the effectiveness of our framework.

Genetic linkage analysis is a recombinant technology used for mapping disease genes on the genome, based on genotypic and phenotypic data collected from families that have affected members. The LOD score is a commonly used statistic in genetic linkage analysis. LOD scores are computed assuming specific values for genetic parameters. However, for complex disorders the specified parameter values are often unknown. One way to address this issue is to maximize the LOD score over all genetic parameters to get a maximum LOD score, or MOD score. Another way is to integrate the LOD score across the genetic parameters to form a posterior probability of linkage, or PPL. Both methods require calculation of large numbers of LOD scores under different sets of parameter values. These calculations may be very time-consuming and can form a significant bottleneck in disease gene mapping. The motivation for this work is to speed up the computation of large numbers of LOD scores in linkage analysis. Instead of the usual LOD calculation where the likelihood of a pedigree under each set of parameter values is computed based on traversing the pedigree, the likelihood of the pedigree is computed here as an algebraic expression that can be optimized and reused. This optimized likelihood expression can be evaluated an arbitrary number of times for LOD scores under different values of the genetic parameters, resulting in much faster speeds. Our initial results show that this approach can speed up the traditional genetic linkage computation by 10~1200 times.

This paper presents the Environment Description Framework (EDF) for modeling complex networks of intersecting roads and pathways in virtual environments. EDF represents information about the layout of streets and sidewalks, the rules that govern behavior on roads and walkways, and the locations of agents with respect to navigable structures. The framework serves as the substrate on which behavior programs for autonomous vehicles and pedestrians are built. Pathways are modeled as ribbons in space. The ribbon structure provides a natural coordinate frame for defining the local geometry of navigable surfaces. EDF includes a powerful runtime interface supported by robust and efficient code for locating objects on the ribbon network, for mapping between Cartesian and ribbon coordinates, and for determining behavioral constraints imposed by the environment.

This paper presents steering behaviors that control autonomous vehicles populating roadways in virtual urban environments. Behavior programming is facilitated by a set of representations of the environment that use convenient frames of reference in natural coordinate systems. Roadway surfaces are modeled as three-dimensional ribbons that make the local orientation of the road explicit and allow relative distances on the road to be simply computed. Roads and intersections are connected to form a ribbon network. An egocentric representation called a path melds road and intersection segments into a single, continuous ribbon that captures the vehicle’s shortterm plan of navigation. A topological structure called a route supports wayfinding. We describe how the interrelated ribbon, path, and route representations are used to build multi-component behaviors that plan routes and safely navigate through traffic filled road networks – tracking lanes, shifting lanes to avoid congestion, anticipating lane changes needed to make turns dictated by the route, negotiating intersections, and respecting the rules of the road.

The paper presents a real-time database for modeling complex networks of intersecting roads and walkways in urban virtual environments. The database represents information about the layout of streets and sidewalks, the rules that govern behavior on roads and walkways, and the locations of agents with respect to road and sidewalk structures. This information is used by programs that control the behavior of autonomous vehicles and pedestrians populating the virtual urban environment. Roads and sidewalks are modeled as ribbons in space. The ribbon structure provides a natural coordinate frame for defining the local geometry of navigable surfaces. This geometry is important for way finding and also forms the geometric basis on which spatial relationships among agents are defined. The database includes a powerful run-time interface supported by robust and efficient code for locating objects on the ribbon network, for mapping between Cartesian and ribbon coordinates, and for determining behavioral constraints imposed by the environment.