In light of the significance of regulatory authorities and the rising demand for information disclosure, a vast amount of information on food safety news reports is readily accessible on the Internet. The extraction of such information for precise classification and provision of appropriate safety alerts based on their respective categories has emerged as a challenging problem for academic research. Given that most food safety-related events in news reports comprise lengthy text, the pre-trained language models currently employed for text analysis are generally limited in their capability to handle long documents. This paper proposes a long-text classification model utilising hierarchical Transformers. We categorise information in long documents into two distinct types: (1) multiple text chunks meeting the length constraint and (2) essential sentences within long documents, such as headings, paragraph start and end sentences, etc. Initially, our proposed model utilises the text chunks as input to the BERT model. Then, it concatenates the output of the BERT model with the important sentences from the document and use them as input to the Transformer model for feature transformation. Finally, we utilise a classifier for food safety news classification. We conducted several comparative experiments with various commonly used text classification models on a dataset constructed from publicly available information on food regulatory websites. Our proposed method outperforms existing methods, establishing itself as the leading approach in terms of performance.

Affective understanding of language is an important research focus in artificial intelligence. The large-scale annotated datasets of Chinese textual affective structure (CTAS) are the foundation for subsequent higher-level analysis of documents. However, there are very few published datasets for CTAS. This paper introduces a new benchmark dataset for the task of CTAS to promote development in this research direction. Specifically, our benchmark is a CTAS dataset with the following advantages: (a) it is Weibo-based, which is the most popular Chinese social media platform used by the public to express their opinions; (b) it includes the most comprehensive affective structure labels at present; and (c) we propose a maximum entropy Markov model that incorporates neural network features and experimentally demonstrate that it outperforms the two baseline models.

We address and formalise the task of sequence-to-sequence (seq2seq) cross-modal retrieval. Given a sequence of text passages as query, the goal is to retrieve a sequence of images that best describes and aligns with the query. This new task extends the traditional cross-modal retrieval, where each image-text pair is treated independently ignoring broader context. We propose a novel variational recurrent seq2seq (VRSS) retrieval model for this seq2seq task. Unlike most cross-modal methods, we generate an image vector corresponding to the latent topic obtained from combining the text semantics and context. This synthetic image embedding point associated with every text embedding point can then be employed for either image generation or image retrieval as desired. We evaluate the model for the application of stepwise illustration of recipes, where a sequence of relevant images are retrieved to best match the steps described in the text. To this end, we build and release a new Stepwise Recipe dataset for research purposes, containing 10K recipes (sequences of image-text pairs) having a total of 67K image-text pairs. To our knowledge, it is the first publicly available dataset to offer rich semantic descriptions in a focused category such as food or recipes. Our model is shown to outperform several competitive and relevant baselines in the experiments. We also provide qualitative analysis of how semantically meaningful the results produced by our model are through human evaluation and comparison with relevant existing methods.

Compared with ordinary news, fake news has the characteristics of faster dissemination, lower production cost and causing great social harm, so the question of how to efficiently and accurately detect fake news has attracted a lot of interest from researchers. We propose a Two-Round Inconsistencybased Multi-modal fusion Network (TRIMOON) for fake news detection, which consists of three main components: the multi-modal feature extraction module, the multi-modal feature fusion module and the classification module. To filter the noise generated in the fusion process, we perform an inconsistency detection not only before, but also after the fusion process, and the experimental results prove this to be effective. Our proposed TRIMOON is evaluated on both the Chinese and the English datasets, and our model outperforms the state-of-the-art approach on several classification evaluation metrics.