Knowledge-Driven Event Embedding for Stock Prediction

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 2133–2142, Osaka, Japan, December 11-17 2016. Knowledge-Driven Event Embedding for Stock Prediction Xiao Ding†∗, Yue Zhang‡, Ting Liu†, Junwen Duan† †Research Center for Social Computing and Information Retrieval Harbin Institute of Technology, China {xding, tliu, jwduan}@ir.hit.edu.cn ‡Singapore University of Technology and Design yue zhang@sutd.edu.sg Abstract Representing structured events as vectors in continuous space offers a new way for defining dense features for natural language processing (NLP) applications. Prior work has proposed effective methods to learn event representations that can capture syntactic and semantic information over text corpus, demonstrating their effectiveness for downstream tasks such as event-driven stock prediction. On the other hand, events extracted from raw texts do not contain background knowl- edge on entities and relations that they are mentioned. To address this issue, this paper proposes to leverage extra information from knowledge graph, which provides ground truth such as at- tributes and properties of entities and encodes valuable relations between entities. Specifically, we propose a joint model to combine knowledge graph information into the objective function of an event embedding learning model. Experiments on event similarity and stock market pre- diction show that our model is more capable of obtaining better event embeddings and making more accurate prediction on stock market volatilities. 1 Introduction Text mining techniques have been used to perform event-driven stock prediction (Ding et al., 2015). The main idea is to learn distributed representations of structured events (i.e. event embeddings) from text, and use them as the basis to generate textual features for predicting price movements in stock markets. Here the definition of events follows the open information extraction literature (Fader et al., 2011; Yates et al., 2007), which has seen applications in semantic parsing (Berant et al., 2013), information retrieval (Sun et al., 2015) and text mining (Ding et al., 2014). Formally, an event is defined as a tuple (A, P, O), where A represents the agent, P represents the predicate and O represents the object. For example, “Mi- crosoft profit rises 11 percent” can be represented as the event tuple (A =“Microsoft profit”, P =“rises”, O =“11 percent”). In addition, the main advantages of event embeddings include (1) they can capture both the syntactic and the semantic information among events and (2) they can be used to alleviate the sparsity of discrete events compared with one-hot feature vectors. The learning principle is that events are syntactically or semantically similar should have similar vectors. The event embedding method of Ding et al. (2015) is based on word embeddings of the agent, pred- icate and object of an event. Neural tensor networks are used to combine the embeddings of the three components into embedding vectors of events. For training, one component of gold-standard event is randomly flipped to synthesize negative examples. This form of event embedding method suffers from some limitations. First, the obtained event embeddings cannot capture the relationship between two syn- tactically or semantically similar events if they do …

Content truncated to 500 words