Text mining of news-headlines for FOREX market prediction

FOREX prediction through text mining of news is viable and effective.Feature-selection by abstraction of word-hypernyms increases prediction accuracy.Feature-weighting based on the sum of pos and neg sentiment scores is effective.Feature-reduction based on maximum optimization for prediction-target is crucial. In this paper a novel approach is proposed to predict intraday directional-movements of a currency-pair in the foreign exchange market based on the text of breaking financial news-headlines. The motivation behind this work is twofold: First, although market-prediction through text-mining is shown to be a promising area of work in the literature, the text-mining approaches utilized in it at this stage are not much beyond basic ones as it is still an emerging field. This work is an effort to put more emphasis on the text-mining methods and tackle some specific aspects thereof that are weak in previous works, namely: the problem of high dimensionality as well as the problem of ignoring sentiment and semantics in dealing with textual language. This research assumes that addressing these aspects of text-mining have an impact on the quality of the achieved results. The proposed system proves this assumption to be right. The second part of the motivation is to research a specific market, namely, the foreign exchange market, which seems not to have been researched in the previous works based on predictive text-mining. Therefore, results of this work also successfully demonstrate a predictive relationship between this specific market-type and the textual data of news. Besides the above two main components of the motivation, there are other specific aspects that make the setup of the proposed system and the conducted experiment unique, for example, the use of news article-headlines only and not news article-bodies, which enables usage of short pieces of text rather than long ones; or the use of general financial breaking news without any further filtration.In order to accomplish the above, this work produces a multi-layer algorithm that tackles each of the mentioned aspects of the text-mining problem at a designated layer. The first layer is termed the Semantic Abstraction Layer and addresses the problem of co-reference in text mining that is contributing to sparsity. Co-reference occurs when two or more words in a text corpus refer to the same concept. This work produces a custom approach by the name of Heuristic-Hypernyms Feature-Selection which creates a way to recognize words with the same parent-word to be regarded as one entity. As a result, prediction accuracy increases significantly at this layer which is attributed to appropriate noise-reduction from the feature-space.The second layer is termed Sentiment Integration Layer, which integrates sentiment analysis capability into the algorithm by proposing a sentiment weight by the name of SumScore that reflects investors' sentiment. Additionally, this layer reduces the dimensions by eliminating those that are of zero value in terms of sentiment and thereby improves prediction accuracy.The third layer encompasses a dynamic model creation algorithm, termed Synchronous Targeted Feature Reduction (STFR). It is suitable for the challenge at hand whereby the mining of a stream of text is concerned. It updates the models with the most recent information available and, more importantly, it ensures that the dimensions are reduced to the absolute minimum.The algorithm and each of its layers are extensively evaluated using real market data and news content across multiple years and have proven to be solid and superior to any other comparable solution. The proposed techniques implemented in the system, result in significantly high directional-accuracies of up to 83.33%.On top of a well-rounded multifaceted algorithm, this work contributes a much needed research framework for this context with a test-bed of data that must make future research endeavors more convenient. The produced algorithm is scalable and its modular design allows improvement in each of its layers in future research. This paper provides ample details to reproduce the entire system and the conducted experiments.