Umeå University's logo

In today's world, following news is crucial for decision-making and staying informed. With the growing volume of daily news, automated processing is essential for timely insights and in aiding individuals and corporations in navigating the complexities of the information society. Another use of automated processing is contextual advertising, which addresses privacy concerns associated with cookie-based advertising by placing ads solely based on web page content, without tracking users or their online behavior. Therefore, accurately determining and categorizing page content is crucial for effective ad placements. The news media, heavily reliant on advertising to sustain operations, represent a substantial market for contextual advertising strategies.

Inspired by these practical applications and the advancements in deep learning over the past decade, this thesis mainly focuses on using deep learning for categorizing news articles into topics of varying granularity. Considering the dynamic nature of these applications and the limited availability of relevant labeled datasets for training models, the thesis emphasizes developing methods that can be trained effectively using unlabeled or partially labeled data. It proposes semi-supervised text classification models for categorizing datasets into predefined coarse-grained topics, where only a few labeled examples exist for each topic, while the majority of the dataset remains unlabeled. Furthermore, to better explore coarse-grained topics within news archives and streams and overcome the limitations of predefined topics in text classification the thesis suggests deep clustering approaches that can be trained in unsupervised settings. 

Moreover, to address the identification of fine-grained topics, the thesis introduces a novel story discovery model for monitoring event-based topics in multi-source news streams. Given that online news reporting often incorporates diverse modalities like text, images, video, and audio to convey information, the thesis finally initiates an investigation into the synergy between textual and visual elements in news article analysis. To achieve this objective, a text-image dataset was annotated, and a baseline was established for event-topic discovery in multimodal news streams. While primarily intended for news monitoring and contextual advertising, the proposed models can, more generally, be regarded as novel approaches in semi-supervised text classification, deep clustering, and news story discovery. Comparison with state-of-the-art baseline models demonstrates their effectiveness in addressing the respective objectives.

Given the importance of identifying and monitoring news stories within the continuous flow of news articles, this paper presents PromptStream, a novel method for unsupervised news story discovery. In order to identify coherent and comprehensive stories across the stream, it is crucial to create article representations that incorporate as much topic-related information from the articles as possible. PromptStream constructs these article embeddings using cloze-style prompting. These representations continually adjust to the evolving context of the news stream through self-supervised learning, employing a contrastive loss and a memory of the most confident article-story assignments from the most recent days. Extensive experiments with real news datasets highlight the notable performance of our model, establishing a new state of the art. Additionally, we delve into selected news stories to reveal how the model’s structuring of the article stream aligns with story progression.

We introduce ADCluster, a deep document clustering approach based on language models that is trained to adapt to the clustering task. This adaptability is achieved through an iterative process where K-Means clustering is applied to the dataset, followed by iteratively training a deep classifier with generated pseudo-labels – an approach referred to as inner adaptation. The model is also able to adapt to changes in the data as new documents are added to the document collection. The latter type of adaptation, outer adaptation, is obtained by resuming the inner adaptation when a new chunk of documents has arrived. We explore two outer adaptation strategies, namely accumulative adaptation (training is resumed on the accumulated set of all documents) and non-accumulative adaptation (training is resumed using only the new chunk of data). We show that ADCluster outperforms established document clustering techniques on medium and long-text documents by a large margin. Additionally, our approach outperforms well-established baseline methods under both the accumulative and non-accumulative outer adaptation scenarios.

We propose a semi-supervised learning method called Cformer for automatic clustering of text documents in cases where clusters are described by a small number of labeled examples, while the majority of training examples are unlabeled. We motivate this setting with an application in contextual programmatic advertising, a type of content placement on news pages that does not exploit personal information about visitors but relies on the availability of a high-quality clustering computed on the basis of a small number of labeled samples.

To enable text clustering with little training data, Cformer leverages the teacher-student architecture of Meta Pseudo Labels. In addition to unlabeled data, Cformer uses a small amount of labeled data to describe the clusters aimed at. Our experimental results confirm that the performance of the proposed model improves the state-of-the-art if a reasonable amount of labeled data is available. The models are comparatively small and suitable for deployment in constrained environments with limited computing resources. The source code is available at https://github.com/Aha6988/Cformer.

We propose a text clustering algorithm that applies an attention mechanism on both word andsentence level. This ongoing work is motivated by an application in contextual programmatic advertising, where the goal is to grouponline articles into clusters corresponding to agiven set of marketing objectives. The maincontribution is the use of attention to identify words and sentences that are of specific importance for the formation of the clusters