| { | |
| "paper_id": "2020", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T10:23:58.234641Z" | |
| }, | |
| "title": "Information Extraction from Federal Open Market Committee Statements", | |
| "authors": [ | |
| { | |
| "first": "Oana", | |
| "middle": [], | |
| "last": "Frunza", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "[email protected]" | |
| }, | |
| { | |
| "first": "Morgan", | |
| "middle": [], | |
| "last": "Stanley", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "We present a novel approach to unsupervised information extraction by identifying and extracting relevant concept-value pairs from textual data. The system's building blocks are domain agnostic, making it universally applicable. In this paper, we describe each component of the system and how it extracts relevant economic information from U.S. Federal Open Market Committee 1 (FOMC) statements. Our methodology achieves an impressive 96% accuracy for identifying relevant information for a set of seven economic indicators: household spending, inflation, unemployment, economic activity, fixed investment, federal funds rate, and labor market.", | |
| "pdf_parse": { | |
| "paper_id": "2020", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "We present a novel approach to unsupervised information extraction by identifying and extracting relevant concept-value pairs from textual data. The system's building blocks are domain agnostic, making it universally applicable. In this paper, we describe each component of the system and how it extracts relevant economic information from U.S. Federal Open Market Committee 1 (FOMC) statements. Our methodology achieves an impressive 96% accuracy for identifying relevant information for a set of seven economic indicators: household spending, inflation, unemployment, economic activity, fixed investment, federal funds rate, and labor market.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "While there are many information extraction approaches described in the literature (Niklaus et al. 2018; Jeanhee and Shawn, 2005) , there are few that can extract targeted concept-value pairs with sentiment scores and are universally applicable to various data types. In addition, many systems require training data or additional corpora to draw meaningful statistics from. We built a framework that does not make use of predefined language patterns or any additional training data to extract targeted concept-value pairs with sentiment scores, across various data types. In addition, the framework provides a means to derive a semantic representation of unstructured data along with supporting sentiment analysis. We applied it to extract a set of economic concepts and their associated values in FOMC statements.", | |
| "cite_spans": [ | |
| { | |
| "start": 83, | |
| "end": 104, | |
| "text": "(Niklaus et al. 2018;", | |
| "ref_id": "BIBREF10" | |
| }, | |
| { | |
| "start": 105, | |
| "end": 129, | |
| "text": "Jeanhee and Shawn, 2005)", | |
| "ref_id": "BIBREF1" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Financial events and information relevant to stakeholders are disseminated at constantly, yielding massive amounts of textual data. In recent years, unstructured textual data has become an important source of information for various financial systems, (Xing et al. 2018) . While some textual data is created at high frequencies, e.g., news headlines, others as-necessary, e.g. SEC 2 filings, others are generated at pre-defined time intervals, e.g. FOMC statements.", | |
| "cite_spans": [ | |
| { | |
| "start": 252, | |
| "end": 270, | |
| "text": "(Xing et al. 2018)", | |
| "ref_id": "BIBREF15" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "The statements appear roughly every 6 weeks but if special circumstances arise, like the Covid-19 pandemic at the beginning of 2020, additional statements are released accordingly. Each FOMC statement lays out the direction of U.S. monetary policy. These statements include the federal funds rate, one of the most important economic indicators. In addition, they contain information about inflation, unemployment, economic growth, and monetary supply. The language used in these statements is carefully selected; every word included is intentional due to the market moving effects of these statements.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Changes in wordings have shown to bring major effects on markets (Gurkaynak et al., 2005) and directions of market movements have shown to be correlated with statement sentiment scores (Rosa, 2011) .", | |
| "cite_spans": [ | |
| { | |
| "start": 65, | |
| "end": 89, | |
| "text": "(Gurkaynak et al., 2005)", | |
| "ref_id": "BIBREF4" | |
| }, | |
| { | |
| "start": 185, | |
| "end": 197, | |
| "text": "(Rosa, 2011)", | |
| "ref_id": "BIBREF11" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "In essence, the most important economic aspects in the statement are federal funds rate, household spending, inflation, unemployment, economic activity, fixed investment, and labor market conditions. Identifying and extracting these concepts along with what is said about them would not only provide a semantic representation of the statement but give rise to a fast integration, as soon as release time, of crucial economic data. Results can be easily and timely integrated into decision making processes, without any human effort. They can be used to derive short-and long-term interest rates, foreign exchange rates, employment rates, economic output, and the supply of credit and demand for investment.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Furthermore, providing sentiment analysis for each economic concept will make the extracted information more valuable to both humans and financial systems.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "In the next sections we are going to present our information extraction system, each of its components, and how we applied it to extract economic concept-value pairs from FOMC statements.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "In this study we propose an unsupervised information extraction framework that is able to extract concept-value pairs from textual data. In addition, our approach differentiates itself from other methods in that it also assigns a sentiment score to each extracted pair. Unsupervised extraction refers to the fact that no training set was provided to the system at the time it was built.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Problem Statement", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "We built this system to extract seven economic concepts and their values from FOMC statements. While the framework is universal and can be applied to any other unstructured textual data without further tuning, we anchor the system's description around FOMC statements. To show that we can apply the framework to other types of data, in the Results section we present system outputs on a sample of short news articles. Table 1 provides an example of an expected output for the information extracted.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 418, | |
| "end": 425, | |
| "text": "Table 1", | |
| "ref_id": "TABREF0" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Problem Statement", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "Expected output household spending: have moderated inflation: 2 percent unemployment: has stayed low economic activity: has been rising at a moderate rate fixed investment: have moderated from their strong fourth-quarter readings federal funds rate: 1-1/2 to 1-3/4 percent labor market: has continued to strengthen Investigating the expected output in Table 1 , the problem we address is a concept-value or key-value pair extraction task. One important aspect to mention is that while we know that we aim to extract values for a targeted set of economic factors it is not necessarily true that the exact factor e.g. \"economic activity\" is present in the statement. More so, when the concept is present it can appear in many places and the system must choose which occurrence to output. As an example, there are many places where \"inflation\" is mentioned in the above statement, \"survey-based measures of longer-term inflation expectations are little changed, on balance\", \"but the Committee is monitoring inflation developments closely\", \"Inflation on a 12-month basis is expected to move up\u2026\" etc., but the only mention that is of interest is the one that contains the rate, \"have continued to run below 2 percent\".", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 352, | |
| "end": 359, | |
| "text": "Table 1", | |
| "ref_id": "TABREF0" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "FOMC statement", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "As mentioned earlier, our approach does not stop at just identifying and selecting key-value pairs. It also assigns a sentiment score to the pair. The sentiment score reflects the tone of the language used to describe the factor, either positive or negative with values between -1 and 1. The sentiment does not take into account the financial aspect but is lexically driven. For example, \"unemployment: has stayed low\", has a negative score since the language used to describe the factor uses words associate with a negative sentiment, even though economically speaking, a low unemployment rate is a positive economic outcome. However, this can be easily remedied by flagging the factors for which the sentiment scores need to be inversed. As a result, the score of the above example will change polarity to become positive.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "FOMC statement", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "In summary, we tackle a concept-value pair information extraction task with sentiment analysis. We focus the task on FOMC statements but build a universal solution for any unstructured textual data.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "FOMC statement", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "The following sections describe the building blocks that constitute our methodology.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "FOMC statement", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "While there is a rich body of literature on information extraction, (Saidul and Vincent Ng, 2014; Chung and Murphy, 2005) there are not many studies that address a concept-value pair extraction task with sentiment analysis for financial text. Sentiment analysis for financial data has been a hot research interest for both natural language processing and finance fields. There is a large body of literature on how to perform this task with solutions falling in two main categories: lexical resource-based, (Loughran and McDonald, 2011; Ke et al., 2019) and machine-learning based, (Zhang et al., 2018) . The financial domain tends to prefer the former mainly due to ease of use and efficiency, (Kearney and Liu, 2014) . Our methodology, unlike related research, makes use of the extraction layer to derive better sentiment scores. We have tried off-the shelf solutions for adding sentiment scores but none of the options were acceptable, (Tang et al., 2009) . A large majority of off-the-shelf solutions are built on product reviews, making them unsuitable for financial data, mostly due to poor coverage and resulting neutral sentiment scores. The literature, mostly on identification of features of product reviews (Zhang and Liu, 2014) makes use of similar building blocks, but our methodology does not heavily rely on problem constraints e.g., use of opinion or sentiment words for aspect identification or it does not yield as comprehensive output as ours. Also, systems that return sentiment scores for very short texts, even for financial data are not widely supported. While we make use of a pre-existing lexical resource, we enhance it with a graphbased methodology to weigh in the importance of each word a technique we have not seen previously. Therefore, the end-to-end architecture that we propose becomes a complete coherent solution.", | |
| "cite_spans": [ | |
| { | |
| "start": 68, | |
| "end": 97, | |
| "text": "(Saidul and Vincent Ng, 2014;", | |
| "ref_id": "BIBREF5" | |
| }, | |
| { | |
| "start": 98, | |
| "end": 121, | |
| "text": "Chung and Murphy, 2005)", | |
| "ref_id": "BIBREF3" | |
| }, | |
| { | |
| "start": 506, | |
| "end": 535, | |
| "text": "(Loughran and McDonald, 2011;", | |
| "ref_id": "BIBREF7" | |
| }, | |
| { | |
| "start": 536, | |
| "end": 552, | |
| "text": "Ke et al., 2019)", | |
| "ref_id": "BIBREF13" | |
| }, | |
| { | |
| "start": 581, | |
| "end": 601, | |
| "text": "(Zhang et al., 2018)", | |
| "ref_id": "BIBREF17" | |
| }, | |
| { | |
| "start": 694, | |
| "end": 717, | |
| "text": "(Kearney and Liu, 2014)", | |
| "ref_id": "BIBREF6" | |
| }, | |
| { | |
| "start": 938, | |
| "end": 957, | |
| "text": "(Tang et al., 2009)", | |
| "ref_id": "BIBREF12" | |
| }, | |
| { | |
| "start": 1217, | |
| "end": 1238, | |
| "text": "(Zhang and Liu, 2014)", | |
| "ref_id": "BIBREF16" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Related Work", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "In terms of FOMC sentiment analysis our method stands out in comparison to other research for the fact that we assign a sentiment score to individual economic concepts rather than to the statement as a whole (Cannon. 2015) . We provide individual sentiment scores for the most salient economic aspects mentioned in the statement, giving users more powerful decision tools.", | |
| "cite_spans": [ | |
| { | |
| "start": 208, | |
| "end": 222, | |
| "text": "(Cannon. 2015)", | |
| "ref_id": "BIBREF0" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Related Work", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "While there is a considerable body of literature around FOMC analysis, a substantial amount is focused on mining meeting notes rather than the statement itself, e.g., identify covered topics in minutes (Hansen et al., 2015) and predict economic indicators like equity and interest rate volatilities (Boukus and Rosenberg, 2006) . Compared to other studies on FOMC statement analysis, our solution brings improvements in terms of both methodology and granularity of the information it extracts.", | |
| "cite_spans": [ | |
| { | |
| "start": 299, | |
| "end": 327, | |
| "text": "(Boukus and Rosenberg, 2006)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Related Work", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "To this point, we have laid the groundwork of the solution proposed and the problem tackled. In this section, we are going to detail our proposed methodology and each building block it contains. Our concept-value pair information extraction system is rooted in dependency parsing outcome, with each subsequent step making use of the previous one, all leading up to a sentiment analysis and a concept-value pair filtering step. In a nutshell, the main components our framework makes use of are as follows:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Methodology", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "1. Dependency parsing 2. Concept-value pair extraction 3. Sentiment analysis 4. Concept-value pair filtering While the end results are relevant concept-value pairs, the input data for our system is the .html file of the statement. The textual body is extracted using the BeautifulSoup 3 package. Figure 1 depicts an overview of our proposed methodology. Figure 1 . Methodology overview.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 296, | |
| "end": 304, | |
| "text": "Figure 1", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 354, | |
| "end": 362, | |
| "text": "Figure 1", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Methodology", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "In the following subsections we describe each individual component of the system.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Methodology", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "While we build our framework on top of a few textual analysis components, syntactic dependency information represents the central part of the methodology we propose. We decided to do this because in the absence of any labeled data, syntactic information is a rich source of knowledge. Also, the language used is grammatically correct resulting in highly accurate parsed data. Finally, sentences tend to be short and factual, allowing for important information to be captured by primary dependency relations.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Dependency parsing information and concept-value pair extraction", | |
| "sec_num": "4.1" | |
| }, | |
| { | |
| "text": "We obtain dependency parsing information by using the Stanford CoreNLP 4 engine on the entire body of a statement. After we retrieve parsed data, the system begins to build concept-value pairs by identifying each \"nsubj\" relation. With all relations identified, the next steps focus on adding targeted dependents, in a recursive manner, to each component of the relation. The below steps highlight the way our methodology makes uses of dependency information:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Dependency parsing information and concept-value pair extraction", | |
| "sec_num": "4.1" | |
| }, | |
| { | |
| "text": "1. Run each sentence of the statement through the parser. 2. Identify subject-verb relation, \"nsubj\", where POS=NN*.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Dependency parsing information and concept-value pair extraction", | |
| "sec_num": "4.1" | |
| }, | |
| { | |
| "text": "3. For both Subject and Governor entities in the \"nsubj\" relation: a. Attach right dependents, if their syntactic relation is part of a pre-defined set. b. Attach left dependents, if their syntactic relation is part of a pre-defined set. c. If the POS of any added dependents is part of a pre-defined set, re-run the steps in 3. There are few pre-defined sets of syntactic information we mention in the above algorithm, syntactic relations for left and right dependents, these are two different sets, and POS information for when to recursively call the algorithm. While our method is unsupervised, the sets of syntactic relations mentioned above were fine-tuned from a process that manually analyzed system results on 3 years of FOMC statements, 2013 through 2015, a total of around 15 statements. We started with lists derived by using the meaning of the syntactic relations then adjusted them by analyzing the results. Figure 2 shows the parse result of a sentence, the starting point of the process. The outcome of running the algorithm consists of a list of concept-value pairs, rooted in nsubj relations. The key is the Subject with its chosen dependents and the value is the Governor with its dependents. For a better understanding of the pair extraction step, we added an example bellow. The concept-pair extraction steps are as follows: 1. Subject relations: 1. \"inflation\" -\"continued\" 2. \"inflation\" -\"continued\" 2. Subject relations dependents:", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 922, | |
| "end": 930, | |
| "text": "Figure 2", | |
| "ref_id": "FIGREF0" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Dependency parsing information and concept-value pair extraction", | |
| "sec_num": "4.1" | |
| }, | |
| { | |
| "text": "1. \"inflation\" -\"continued\" 1. \"inflation\" -> 1. both overall inflation and inflation for items -have continued to run below 2 percent 2. inflation -have continued to run below 2 percent", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Dependency parsing information and concept-value pair extraction", | |
| "sec_num": "4.1" | |
| }, | |
| { | |
| "text": "Subsequent to the dependency parsing module is the sentiment analysis step. Each concept-pair extracted in the previous step will receive a sentiment score.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Sentiment analysis", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "Since we tried off-the-shelf tools, without being able to successfully capture the sentiment of the language used in the statements, the vast majority of scores turn neutral, we decided to perform this step with a custom-based solution. In fact, it is well known that sentiment analysis tools that are trained on customer product reviews, like the vast majority of available tools, are not suitable for financial texts. Loughran and McDonald (2011) found out that a large majority of negative words aren't negative if used in financial contexts. As an example, the word \"bond\" in non-financial data carries a positive connotation, but in finance is just a financial instrument that carries an objective charge.", | |
| "cite_spans": [ | |
| { | |
| "start": 420, | |
| "end": 448, | |
| "text": "Loughran and McDonald (2011)", | |
| "ref_id": "BIBREF7" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Sentiment analysis", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "Our proposed solution is rooted in SentiWordNet, (Esuli Andrea and Sebastiani, 2010) a lexical resource based on WordNet 5 . In SentiWordNet, each word is associated with three scores: objective, positive, and negative. These scores are obtained using a statistical model. We decided to use this resource to leverage the lexical and semantic nature of WordNet and the part-of-speech information. It provided superior results even though it is not specifically tuned for financial data. It also allowed us to derive custom sentiment scores and to easily and efficiently perform a sentiment analysis step.", | |
| "cite_spans": [ | |
| { | |
| "start": 49, | |
| "end": 84, | |
| "text": "(Esuli Andrea and Sebastiani, 2010)", | |
| "ref_id": "BIBREF2" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Sentiment analysis", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "To make our sentiment score more meaningful we accounted for, only the words that belong to the top key phrases extracted by the TextRank algorithm, (Mihalcea and Tarau, 2004) . The algorithm identifies the most central key phrases in the text by using a rank-based graph method.", | |
| "cite_spans": [ | |
| { | |
| "start": 149, | |
| "end": 175, | |
| "text": "(Mihalcea and Tarau, 2004)", | |
| "ref_id": "BIBREF9" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Sentiment analysis", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "Given the above observations around our sentiment analysis component, the approach we implement to derive a sentiment score for the value of each concept-pair is as follows:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Sentiment analysis", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "1. Run the TextRank algorithm on the statement to identify top keyphrases 2. For each value word that belongs to the selected keyphrases, calculate the (positive -negative) SentiWordNet score, taking into account POS information. a. If negation present, negate value. 3. Compute concept-value pair sentiment score as the average of the scores from step 2. Using the approach described above, below are some examples of assigned scores:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Sentiment analysis", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "\uf0b7 advanced: 0.1406 \uf0b7 remains elevated: 0.0729 \uf0b7 continued to advance: 0.0417 \uf0b7 declined: -0.0893 \uf0b7 has stayed low: -0.2051 From these examples we can observe how our method is able to capture various nuances of the language: \"declined\" has a lower score than \"has stayed low\". At the end of this step, our framework assigns a sentiment score for each extracted pair with values ranging from -1 to 1.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Sentiment analysis", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "The last component we implement is a selection step where we only select one concept-value pair for each of the economic factors of interest from all extracted concept-value pairs. The ultimate goal is to expose the one concept-value pair that captures the right information, as presented in Table 1 .", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 292, | |
| "end": 299, | |
| "text": "Table 1", | |
| "ref_id": "TABREF0" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Concept-value pair filtering", | |
| "sec_num": "4.3" | |
| }, | |
| { | |
| "text": "The way we conduct this phase is by implementing a set of functions that act as adjustable knobs. One function is aimed to filter and identify relevant pairs by searching the economic factor in either the key, the value, or both parts of a pair. Another function supports a search step that allows for both string and embedding-based matching. Here, an embedding space of the implementer's choice can be used.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Concept-value pair filtering", | |
| "sec_num": "4.3" | |
| }, | |
| { | |
| "text": "Besides the availability of robust search strategies, we also defined and implemented a set of targeted constraints a pair needs to follow e.g., should contain numbers, returns only the shortest pair, etc. While the method has some default behavior and returns the best shortest match it also supports the return of the entire set of best matches.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Concept-value pair filtering", | |
| "sec_num": "4.3" | |
| }, | |
| { | |
| "text": "One of the most robust features of the system is the fact that it is very easy to enhance it to target additional economic factors e.g., job status. In order to achieve this, simply name a new factor of interest and if necessary, define which constraints the pairs need to follow.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Concept-value pair filtering", | |
| "sec_num": "4.3" | |
| }, | |
| { | |
| "text": "Below are some examples of extracted pairs that show the language variations the method is able to capture. The text in bold represents the concept part of the pair while the underlined text is the value.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Concept-value pair filtering", | |
| "sec_num": "4.3" | |
| }, | |
| { | |
| "text": "\uf0b7 On a 12-month basis, overall inflation and the measure excluding food and energy prices have declined and are running below 2 percent. \uf0b7 Labor market conditions improved, with the unemployment rate declining further. \uf0b7 Recent data suggest that growth of household spending moderated from its strong fourthquarter pace, while business fixed investment continued to grow strongly. \uf0b7 Household spending has continued to rise moderately while business fixed investment has remained soft. The filtering module is the last step, at the end of which the best concept-value pairs are identified.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Concept-value pair filtering", | |
| "sec_num": "4.3" | |
| }, | |
| { | |
| "text": "For each given statement and for each economic factor, the system will return a single best conceptvalue pair with a sentiment score. When measuring the system's performance on how well it is able to identify and extract the correct pairs, both precision and recall stand at 96%. These results are manually obtained on a dataset of 30 statements of 5 years, 2016 to 2020, inclusive.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Results", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "The cases where the method is not identifying or extracting the correct information are mostly due to one of the following reasons: the quality of the parsing output, search match, e.g. while the method extracts the pair \"investment -advanced more quickly in recent months\" the filtering step fails to select it due to missing string match with \"fixed investment\", and occasionally large variations in the language of the statement that do not allow the method to extract the complete relevant value.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Results", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "The default system output is an .xlsx file that contains the extracted concept-value pairs and their sentiment scores. This format can be easily integrated with automatic systems. In addition, to show how the extracted information can be used for human consumption, Figure 3 presents results obtained over a period of 8 years, 2013 to 2020, for two of the economic factors. We used bars to represent federal-funds rates, one of the most important economic information, and labels to depict economic factors values. We color coded the bars to reflect the sentiment scores, red color encodes a positive sentiment while blue a negative one. Sentiment intensity is reflected in the darkness of the color. In addition, the lower part of the plot captures the sentiment score throughout the years providing users with a historic view of how the sentiment of an economic factor changes and how it is correlated with the federal funds rate.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 266, | |
| "end": 274, | |
| "text": "Figure 3", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Results", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "As we can observe, our method is not only able to identify and surface the right information but also provides a sentiment analysis history for the economic factors. This aspect in itself represents a Figure 3 . System results on FOMC statements. valuable source of information for financial systems that try to predict the type of economy we could expect. More so, the extracted concept-value pairs can provide a semantic representation of an FOMC statement given that it captures the most salient information. Revealing such a semantically charged representation allows both humans and financial systems to easily compare and contrast how various economic aspects change throughout the years.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 201, | |
| "end": 209, | |
| "text": "Figure 3", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Results", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "For a more complete analysis we run our methodology on news articles. Figure 4 presents results on the following news body \"The U.S. dollar declined against most major foreign currencies yesterday, although the drop was softened when bond prices failed to advance Tuesday's rally. The dollar began weakening in Europe as interest rates fell there for dollar deposits. The decline continued in New York trading, which was thin, although the dollar recovered slightly when bond prices began falling. Lower bond prices translate into higher long-term interest yields, which make dollar denominated investments more attractive. The bond market later closed littlechanged from Tuesday. 'This is the first time in a while that we've gone back to trading off interest rates, and my feeling is it will continue between now and the (U.S.) election,' said Daniel Holland, an assistant vice president at Discount Corp., New York. In late New York trading, the dollar fell to 3.0210 West German marks from 3.0318 marks on Tuesday. The British pound rose to $1.2223 from $1.2155. In early Tokyo trading Thursday, the dollar strengthened against the Japanese currency, to 245.45 yen from 245.13 yen late yesterday in New York.\" 6 Figure 4 . System results on news.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 70, | |
| "end": 78, | |
| "text": "Figure 4", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 1216, | |
| "end": 1224, | |
| "text": "Figure 4", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Results", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "The extracted key-value pairs show that we can easily obtain a semantic summary of what has been mentioned in the body along with the sentiment it conveys. Results support the fact that our methodology extracts important information from a variety of textual data.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Results", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "In this study we presented a novel unsupervised information extraction system with sentiment analysis. Our proposed architecture is robust, as it can be extended to support new targeted concepts, it can easily adapt to user-defined constraints, and it can be tuned to vary the amount of information it returns. We achieve 96% accuracy at extracting relevant concept pairs from FOMC statements and we show that we can use the same methodology on completely different data than the one used for tuning.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusions", | |
| "sec_num": "6" | |
| }, | |
| { | |
| "text": "https://www.crummy.com/software/BeautifulSoup/bs4/doc/ 4 https://stanfordnlp.github.io/CoreNLP/", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "https://wordnet.princeton.edu/", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "https://www.figure-eight.com/data-for-everyone/", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [ | |
| { | |
| "text": "I would like to thank Morgan Stanley head of Transformation Division, MD Sigal Zarmi and the head of the Montreal, Canada office, MD Alan Vesprini for their support.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Acknowledgements", | |
| "sec_num": null | |
| } | |
| ], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "Sentiment of the FOMC: Unscripted. Federal Reserve Bank of Kansas City Economic Review", | |
| "authors": [ | |
| { | |
| "first": "San", | |
| "middle": [], | |
| "last": "Cannon", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2015, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "San Cannon. 2015. Sentiment of the FOMC: Unscripted. Federal Reserve Bank of Kansas City Economic Review (Fourth Quarter).", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "Concept-value pair extraction from semi-structured clinical narrative: a case study using echocardiogram reports", | |
| "authors": [ | |
| { | |
| "first": "Jeanhee", | |
| "middle": [], | |
| "last": "Chung", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Shawn", | |
| "middle": [], | |
| "last": "Murphy", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2005, | |
| "venue": "AMIA Annual Symposium Proceedings", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "131--135", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Jeanhee Chung, Shawn Murphy. 2005. Concept-value pair extraction from semi-structured clinical narrative: a case study using echocardiogram reports. AMIA Annual Symposium Proceedings, 2005:131-135.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Sentiwordnet 3.0: An enhanced lexical resource for sentiment analysis and opinion mining", | |
| "authors": [ | |
| { | |
| "first": "Andrea", | |
| "middle": [], | |
| "last": "Esuli", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Fabrizio", | |
| "middle": [], | |
| "last": "Sebastiani", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2010, | |
| "venue": "Proceedings of the Seventh International Conference on Language Resources and Evaluation (LREC'10)", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "2200--2204", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Andrea Esuli and Fabrizio Sebastiani. 2010. Sentiwordnet 3.0: An enhanced lexical resource for sentiment analysis and opinion mining. In Proceedings of the Seventh International Conference on Language Resources and Eval- uation (LREC'10), pages: 2200-2204, Valletta, Malta.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "Concept-Value Pair Extraction from Semi-Structured Clinical Narrative: A Case Study Using Echocardiogram Reports", | |
| "authors": [ | |
| { | |
| "first": "Chung", | |
| "middle": [], | |
| "last": "Jeanhee", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Murphy", | |
| "middle": [], | |
| "last": "Shawn", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2005, | |
| "venue": "AMIA Annual Symposium proceedings / AMIA Symposium. AMIA Symposium", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "131--136", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Chung Jeanhee and Murphy Shawn. 2005. Concept-Value Pair Extraction from Semi-Structured Clinical Narra- tive: A Case Study Using Echocardiogram Reports. AMIA Annual Symposium proceedings / AMIA Sympo- sium. AMIA Symposium 2005, pages: 131-5.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Do actions speak louder than words? The response of asset prices to monetary policy actions and statements", | |
| "authors": [ | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Refet", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Brian", | |
| "middle": [], | |
| "last": "Gurkaynak", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Eric", | |
| "middle": [ | |
| "T" | |
| ], | |
| "last": "Sack", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Swanson", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2005, | |
| "venue": "Intl J Central Banking", | |
| "volume": "1", | |
| "issue": "1", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Refet S. Gurkaynak, Brian Sack, and Eric T. Swanson. 2005. Do actions speak louder than words? The response of asset prices to monetary policy actions and statements. Intl J Central Banking 1(1). May 2005.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Automatic Keyphrase Extraction: A Survey of the State of the Art", | |
| "authors": [ | |
| { | |
| "first": "Kazi", | |
| "middle": [], | |
| "last": "Hasan", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Vincent", | |
| "middle": [], | |
| "last": "Saidul", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Ng", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1262--1273", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Hasan Kazi Saidul and Vincent Ng. 2014. Automatic Keyphrase Extraction: A Survey of the State of the Art. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, 2014, pages 1262- 1273, Baltimore, Maryland, USA.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "Textual sentiment in finance: A survey of methods and models", | |
| "authors": [ | |
| { | |
| "first": "Colm", | |
| "middle": [], | |
| "last": "Kearney", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Sha", | |
| "middle": [], | |
| "last": "Liu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "International Review of Financial Analysis", | |
| "volume": "33", | |
| "issue": "", | |
| "pages": "171--185", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Colm Kearney and Sha Liu. 2014. Textual sentiment in finance: A survey of methods and models. International Review of Financial Analysis, 33:171-185.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "When a liability is not a liability? Textual analysis, dictionaries, and 10-Ks", | |
| "authors": [ | |
| { | |
| "first": "Tim", | |
| "middle": [], | |
| "last": "Loughran", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Bill", | |
| "middle": [], | |
| "last": "Mcdonald", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "The Journal of Finance", | |
| "volume": "66", | |
| "issue": "1", | |
| "pages": "35--65", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tim Loughran, and Bill McDonald. 2011. When a liability is not a liability? Textual analysis, dictionaries, and 10- Ks. The Journal of Finance 66(1): 35-65.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Good Debt or Bad Debt: Detecting Semantic Orientations in Economic Texts", | |
| "authors": [ | |
| { | |
| "first": "Pekka", | |
| "middle": [], | |
| "last": "Malo", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ankur", | |
| "middle": [], | |
| "last": "Sinha", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Pyry", | |
| "middle": [], | |
| "last": "Takala", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Pekka", | |
| "middle": [], | |
| "last": "Korhonen", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jyrki", | |
| "middle": [], | |
| "last": "Wallenius", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Journal of the American Society for Information Science and Technology", | |
| "volume": "65", | |
| "issue": "", | |
| "pages": "782--796", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Pekka Malo, Ankur Sinha, Pyry Takala, Pekka Korhonen, Jyrki Wallenius. 2014. Good Debt or Bad Debt: Detect- ing Semantic Orientations in Economic Texts. Journal of the American Society for Information Science and Technology, 65, 4, pp. 782-796.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "Textrank: Bringing order into texts", | |
| "authors": [ | |
| { | |
| "first": "Rada", | |
| "middle": [], | |
| "last": "Mihalcea", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Paul", | |
| "middle": [], | |
| "last": "Tarau", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2004, | |
| "venue": "Proceedings of the Conference on Empirical Methods in Natural Language Processing", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "404--411", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Rada Mihalcea and Paul Tarau. 2004. Textrank: Bringing order into texts. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, pages 404-411, Barcelona, Spain.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "A Survey on Open Information Extraction", | |
| "authors": [ | |
| { | |
| "first": "Christina", | |
| "middle": [], | |
| "last": "Niklaus", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Matthias", | |
| "middle": [], | |
| "last": "Cetto", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "Proceedings of the 27th International Conference on Computational Linguistics", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Christina Niklaus, Matthias Cetto, Andr\u00e9 Freitas and Siegfried Handschuh. 2018. A Survey on Open Information Extraction. In Proceedings of the 27th International Conference on Computational Linguistics, pages 3866- 3878, Santa Fe, NM, USA. 33.", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "Words that shake traders: the stock markets reaction to central bank communication in real time", | |
| "authors": [ | |
| { | |
| "first": "Carlo", | |
| "middle": [], | |
| "last": "Rosa", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2011, | |
| "venue": "Journal of Empirical Finance", | |
| "volume": "18", | |
| "issue": "5", | |
| "pages": "915--934", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Carlo Rosa. 2011. Words that shake traders: the stock markets reaction to central bank communication in real time. Journal of Empirical Finance Volume 18, Issue 5, 2011, pages 915-934.", | |
| "links": null | |
| }, | |
| "BIBREF12": { | |
| "ref_id": "b12", | |
| "title": "A survey on sentiment detection of reviews", | |
| "authors": [ | |
| { | |
| "first": "Huifeng", | |
| "middle": [], | |
| "last": "Tang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Songbo", | |
| "middle": [], | |
| "last": "Tan", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Xueqi", | |
| "middle": [], | |
| "last": "Cheng", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2009, | |
| "venue": "Expert Systems with Applications", | |
| "volume": "36", | |
| "issue": "", | |
| "pages": "10760--10773", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Huifeng Tang, Songbo Tan, and Xueqi Cheng. 2009. A survey on sentiment detection of reviews. Expert Systems with Applications (ELSEVIER), 2009, 36:10760-10773.", | |
| "links": null | |
| }, | |
| "BIBREF13": { | |
| "ref_id": "b13", | |
| "title": "Predicting Returns with Text Data", | |
| "authors": [ | |
| { | |
| "first": "Zheng", | |
| "middle": [], | |
| "last": "Tracy Ke", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Bryan", | |
| "middle": [ | |
| "T" | |
| ], | |
| "last": "Kelly", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Dacheng", | |
| "middle": [], | |
| "last": "Xiu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "NBER Working Paper Series", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Zheng Tracy Ke, Bryan T. Kelly, and Dacheng Xiu. 2019. Predicting Returns with Text Data. Cambridge, MA: NBER Working Paper Series.", | |
| "links": null | |
| }, | |
| "BIBREF14": { | |
| "ref_id": "b14", | |
| "title": "Knowledge-Driven Approaches for Financial News Analytics", | |
| "authors": [ | |
| { | |
| "first": "Bikesh", | |
| "middle": [], | |
| "last": "Upreti", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Philipp", | |
| "middle": [], | |
| "last": "Back", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Pekka", | |
| "middle": [], | |
| "last": "Malo", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Oskar", | |
| "middle": [], | |
| "last": "Ahlgren", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Ankur", | |
| "middle": [], | |
| "last": "Sinha", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2019, | |
| "venue": "Network Theory and Agent-Based Modeling in Economics and Finance", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "375--404", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Bikesh Upreti, Philipp Back, Pekka Malo, Oskar Ahlgren, and Ankur Sinha. 2019. Knowledge-Driven Approaches for Financial News Analytics. In: Network Theory and Agent-Based Modeling in Economics and Finance. Springer, Singapore: pages 375-404, Springer, Singapore.", | |
| "links": null | |
| }, | |
| "BIBREF15": { | |
| "ref_id": "b15", | |
| "title": "Natural language based financial forecasting: A survey. Artificial Intelligence Review", | |
| "authors": [ | |
| { | |
| "first": "Frank", | |
| "middle": [], | |
| "last": "Xing", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Erik", | |
| "middle": [], | |
| "last": "Cambria", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Roy", | |
| "middle": [], | |
| "last": "Welsch", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": { | |
| "DOI": [ | |
| "10.1007/s10462-017-9588-9" | |
| ] | |
| }, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Frank Xing, Erik Cambria, and Roy Welsch. 2018. Natural language based financial forecasting: A survey. Artifi- cial Intelligence Review. 2018, doi:10.1007/s10462-017-9588-9.", | |
| "links": null | |
| }, | |
| "BIBREF16": { | |
| "ref_id": "b16", | |
| "title": "Aspect and entity extraction for opinion mining", | |
| "authors": [ | |
| { | |
| "first": "Lei", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Bing", | |
| "middle": [], | |
| "last": "Liu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2014, | |
| "venue": "Data Mining and Knowledge Discovery for Big Data", | |
| "volume": "1", | |
| "issue": "", | |
| "pages": "1--40", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Lei Zhang and Bing Liu. 2014. Aspect and entity extraction for opinion mining. In Data Mining and Knowledge Discovery for Big Data, Volume 1, pages 1-40. Springer.", | |
| "links": null | |
| }, | |
| "BIBREF17": { | |
| "ref_id": "b17", | |
| "title": "Deep learning for sentiment analysis: A survey", | |
| "authors": [ | |
| { | |
| "first": "Lei", | |
| "middle": [], | |
| "last": "Zhang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Shuai", | |
| "middle": [], | |
| "last": "Wang", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Bing", | |
| "middle": [], | |
| "last": "Liu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 2018, | |
| "venue": "Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Lei Zhang, Shuai Wang, and Bing Liu. 2018. Deep learning for sentiment analysis: A survey. In Wiley Interdisci- plinary Reviews: Data Mining and Knowledge Discovery, page e1253. Wiley Online Library.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "FIGREF0": { | |
| "uris": null, | |
| "text": "Dependency parsing output.", | |
| "type_str": "figure", | |
| "num": null | |
| }, | |
| "TABREF0": { | |
| "type_str": "table", | |
| "num": null, | |
| "content": "<table/>", | |
| "html": null, | |
| "text": "Example of an FOMC statement with expected output." | |
| } | |
| } | |
| } | |
| } |