[Paper Review] Sentiment Analysis of German Twitter

Wladimir Sidorenko|arXiv (Cornell University)|Jan 1, 2019

Sentiment Analysis and Opinion Mining155 references4 citations

TL;DR

This paper introduces a large-scale, manually annotated German Twitter sentiment corpus and proposes novel methods for sentiment analysis in German social media. It improves sentiment lexicon generation, fine-grained opinion mining using enhanced CRFs, message-level classification with lexicon-attention mechanisms, and discourse-aware analysis via latent-marginalized CRFs and recursive Dirichlet processes—achieving state-of-the-art performance on German Twitter sentiment tasks.

ABSTRACT

The immense popularity of online communication services in the last decade has not only upended our lives (with news spreading like wildfire on the Web, presidents announcing their decisions on Twitter, and the outcome of political elections being determined on Facebook) but also dramatically increased the amount of data exchanged on these platforms. Therefore, if we wish to understand the needs of modern society better and want to protect it from new threats, we urgently need more robust, higher-quality natural language processing (NLP) applications that can recognize such necessities and menaces automatically, by analyzing uncensored texts. Unfortunately, most NLP programs today have been created for standard language, as we know it from newspapers, or, in the best case, adapted to the specifics of English social media. This thesis reduces the existing deficit by entering the new frontier of German online communication and addressing one of its most prolific forms—users’ conversations on Twitter. In particular, it explores the ways and means by how people express their opinions on this service, examines current approaches to automatic mining of these feelings, and proposes novel methods, which outperform state-of-the-art techniques. For this purpose, I introduce a new corpus of German tweets that have been manually annotated with sentiments, their targets and holders, as well as lexical polarity items and their contextual modifiers. Using these data, I explore four major areas of sentiment research: (i) generation of sentiment lexicons, (ii) fine-grained opinion mining, (iii) message-level polarity classification, and (iv) discourse-aware sentiment analysis. In the first task, I compare three popular groups of lexicon generation methods: dictionary-, corpus-, and word-embedding–based ones, finding that dictionary-based systems generally yield better polarity lists than the last two groups. Apart from this, I propose a linear projection algorithm, whose results surpass many existing automatically-generated lexicons. Afterwords, in the second task, I examine two common approaches to automatic prediction of sentiment spans, their sources, and targets: conditional random fields (CRFs) and recurrent neural networks, obtaining higher scores with the former model and improving these results even further by redefining the structure of CRF graphs. When dealing with message-level polarity classification, I juxtapose three major sentiment paradigms: lexicon-, machine-learning–, and deep-learning–based systems, and try to unite the first and last of these method groups by introducing a bidirectional neural network with lexicon-based attention. Finally, in order to make the new classifier aware of microblogs' discourse structure, I let it separately analyze the elementary discourse units of each tweet and infer the overall polarity of a message from the scores of its EDUs with the help of two new approaches: latent-marginalized CRFs and Recursive Dirichlet Process.

Motivation & Objective

• To address the lack of high-quality, manually annotated German social media data for sentiment analysis.
• To develop and evaluate novel methods for sentiment analysis in German Twitter, focusing on lexicon generation, opinion mining, message-level classification, and discourse-aware analysis.
• To create a comprehensive resource for training and evaluating German NLP systems in low-resource, informal language settings.
• To improve performance on sentiment analysis tasks by integrating linguistic structure, contextual modifiers, and discourse-aware modeling.

Proposed method

• Proposes a new, manually annotated German Twitter corpus with sentiment labels, targets, holders, and lexical polarity items.
• Compares dictionary-, corpus-, and word embedding-based lexicon generation, favoring dictionary-based methods and introducing a linear projection algorithm.
• Uses conditional random fields (CRFs) with restructured graph topologies for improved fine-grained opinion mining.
• Introduces a bidirectional neural network with lexicon-based attention for message-level sentiment classification.
• Applies latent-marginalized CRFs and Recursive Dirichlet Processes to model discourse structure and infer overall tweet polarity from elementary discourse units.
• Employs belief propagation and Viterbi decoding for inference in linear-chain, semi-Markov, and tree-structured CRFs with modified α and β score computations.

Experimental results

Research questions

• RQ1 Which method—dictionary-based, corpus-based, or word embedding-based—yields the most reliable sentiment lexicons for German Twitter?
• RQ2 Can restructured CRF graphs improve the performance of fine-grained opinion mining in German tweets?
• RQ3 How does integrating lexicon-based attention into a bidirectional neural network affect message-level sentiment classification?
• RQ4 To what extent does modeling discourse structure enhance sentiment classification on microblogs?
• RQ5 Can latent-marginalized CRFs and recursive Dirichlet processes effectively model discourse-aware sentiment inference in tweets?

Key findings

• Dictionary-based lexicon generation outperforms corpus- and word embedding-based methods in terms of polarity list quality.
• The proposed linear projection algorithm surpasses many existing automatically generated lexicons in performance.
• CRF-based models with restructured graphs achieve higher scores in fine-grained opinion mining than standard CRFs or RNNs.
• The bidirectional neural network with lexicon-based attention improves message-level sentiment classification by combining lexicon and deep learning strengths.
• Latent-marginalized CRFs and recursive Dirichlet processes enhance discourse-aware sentiment analysis by modeling elementary discourse units and their hierarchical relationships.
• The proposed methods achieve state-of-the-art results across all four sentiment analysis tasks on the new German Twitter corpus.