IRJET- Text based Deep Learning for Stock Prediction
0 likes62 views
This document presents a proposed system for text-based deep learning stock prediction and interpretation. It discusses using a neural network model to extract relevant predictive factors from news texts and financial tweets that influence stock prices. An interactive visualization interface is explored to effectively communicate the interpreted model predictions to users. The system aims to help with stock market investment and analysis tasks. It evaluates the approach on two case studies predicting stock prices from online news and tweets, finding the proposed neural network architecture outperforms other models.
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 05 | May 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 6093
Stock Market Exchange
8. CONCLUSION
This paper proposed to build a stock network from co-
occurrences of ticker symbols in tweets. The properties of
SSN reveal some close relationships between involved
stocks, which provide goodinformationforpredictingstocks
based on social sentiment. Experimental resultsshowedthat
event embeddings-based document representations are
better than discrete events-based methods, and deep
convolutional neural network can capture longer-term
influence of news event than standard feedforward neural
network.
REFERENCES
[1] Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,”
Nature, vol. 521, no. 7553, pp. 436–444, 2015.
[2] M. Liu, J. Shi, Z. Li, C. Li, J. Zhu, and S. Liu, “Towards
better analysis of deep convolutional neural networks,”
IEEE Transactions on Visualization and Computer
Graphics, vol. 23, no. 1, pp. 91–100, 2017.
[3] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual
learning for image recognition,” arXiv:1512.03385,
2015.
[4] K. Simonyan, A. Vedaldi, and A. Zisserman, “Deep inside
convolutional networks: Visualisingimageclassification
models and saliency maps,” arXiv:1312.6034, 2013.
[5] L. M. Zintgraf, T. S. Cohen, and M. Welling, “A new
method to visualize deep neural networks,”
arXiv:1603.02518, 2016.
[6] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imagenet
classification with deep convolutional neural networks,”
in NIPS’12, 2012, pp. 1097– 1105.
[7] M. D. Zeiler and R. Fergus, “Visualizing and
understanding convolutional networks,” in ECCV’14,
2014, pp. 818–833.
[8] D. E. Rumelhart, G. E. Hinton, and R. J. Williams,
“Learning representations by back-propagating errors,”
Nature, vol. 323, no. 6088, pp. 533–536, 1986.
[9] G. Neubig and Others, “DyNet: The dynamic neural
network toolkit,” arXiv preprint arXiv:1701.03980,
2017.
[10] “Deep learning for event-driven stock prediction,” in
IJCAI’15, 2015, pp. 2327–2333. K.](https://image.slidesharecdn.com/irjet-v6i5833-191010085656/85/IRJET-Text-based-Deep-Learning-for-Stock-Prediction-4-320.jpg)
IRJET- Text based Deep Learning for Stock Prediction
- 1. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 05 | May 2019 www.irjet.net p-ISSN: 2395-0072 © 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 6090 Text based Deep Learning for Stock Prediction Miss.Chavan Shamal S1, Miss. Amrutkar Pallavi A2, Miss. Pawar Priyanka B3, Miss. Waman Monali A4 1,2,3,4SND College of Engineering & Research Centre, Yeola ---------------------------------------------------------------------***--------------------------------------------------------------------- Abstract: We propose a deep learning method for event driven stock market prediction. First, events are extracted from news text, and represented as dense vectors, trained using a novel neural tensor network. Second,a deep convolutional neural network is used to model both short- term and long-term influences of events on stock price movements. For example, professional traders in their long- term careers have accumulated numerous trading rules, the myth of which they can understand quite well. On the other hand, deep learning models have been hardly interpretable. This paper presents DeepClue, a system built to bridge text- based deep learning models and end users through visually interpreting the key factors learned in the stock price prediction model. We make three contributionsinDeepClue. First, by designing the deep neural network architecture for interpretation and applying an algorithm to extract relevant predictive factors, we provide a useful case on what can be interpreted out of the prediction model for end users. Second, by exploring hierarchies over the extracted factors and displaying these factors in an interactive, hierarchical visualization interface, we shed light on how to effectively communicate the interpreted model to end users. Third, we evaluate the integrated visualization system through two case studies in predicting the stock price with online financial news and company related tweets from social media. Quantitative experiments comparing the proposed neural network architecture with state-of-the-art models and the human baseline are conducted and reported. All the study results demonstrate the effectiveness of DeepClue in helping to complete stock market investment and analysis tasks. Keywords: Deep learning, visualization, model interpretation, stock prediction, neural network. 1. INTRODUCTION In this paper, we target the research problem of how to interpret text-based deep stock prediction model for end users, so that they can make up their stock trading decisions as well as improve the prediction model based on the interpretation. In particular, we investigate research questions including what kind of information can be effciently extracted from prediction model as interpretations, and how to communicate such information in an effective way to end users. Throughout this work, we depend on an interactive visualizationinterfaceto bridge the prediction model and end users, which turns out a natural and straightforward choice. DEEP learning techniques are reshaping the landscape of predictiveanalysisinthe bigdata research area and have made major breakthroughs inimage and speech recognition, question answering , machine translation and many other application domains. For example, fanancial news such as Amazon port beats forecasts was accompanied with a surge of Amazons stock price, while Oil price hits a record high triggered worries on the auto industry and weakened their performance in the stock market. 2. LITERATURE SURVEY 1. “DeepClue:Visual Interpretation of Text –based Deep Stock Prediction” This paper presents deep clue, a system build to bridge text based deep learning model and end user through visually interpretingthe key factor learninthestock price prediction model. All the studyresultsdemonstratethe effectiveness of DeepClue in helping to complete stock market investment and analysis tasks. 2. “Stock Prediction Using Twitter Sentiment Analysis” In order to test our results, we propose a new cross validation method for financial data and obtain 75.56% accuracy using Self Organizing Fuzzy Neural Networks (SOFNN) on the Twitter feeds and DJIA values from the period June 2009 to December 2009. We also implement a naive protfolio management strategy based on our predicted values. Our work is based on Bollen et als famous paper which predicted the same with 87% accuracy. 3. “Deep Learning for Event-Driven Stock Prediction” We propose a deep learning method for event driven stock market prediction. First, events are extractedfromnewstext, and represented as dense vectors, trained using a novel neural tensor network. Second, a deep convolutional neural network is used to model both short-term and long-term influences of events on stock price movements. 4. “Exploiting Social Relations and Sentiment for Stock Prediction” we first exploit cash-tags (followed by stocks ticker symbols) in Twitter to build a stock network, where nodes are stocks connected by edges when two stocks co occur frequently in tweets. We then employ a labeled topic model to jointly model both the tweets and the network structure to assign each node and each edge a topic respectively.ThisSemanticStockNetwork(SSN)summarizes discussion topicsaboutstocksandstockrelations.Wefurther show that social sentiment about stock (node) topics and stock relationship (edge) topics are predictive of each stocks market.
- 2. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 05 | May 2019 www.irjet.net p-ISSN: 2395-0072 © 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 6091 5. “ImageNet Classification with Deep Convolutional Neural Networks” We trained a large, deep convolutional neural network to classify the 1.2 million high resolution images in the ImageNet LSVRC-2010 contest into the 1000 different classes. On the test data, we achieved top-1 and top-5 error rates of 37.5% and 17.0% which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max- pooling layers, and three fully-connected layers with a final 1000-way softmax. 3. EXISTING SYSTEM Tweets in Relation to the Stock Market : Micro-blogging activities are well correlated with the stock market. Figure 3 shows us how the Twitter activities response to a report announcement of $aapl (Jan. 23 2013). The report was made public soon after the market closed at 4:00pm, while the tweets volume rose about two hours earlier and reached the peak at the time of announcement, then it arrived the second peak at the time near the market’s next opening (9:30am). By further accumulating all days’ tweet volume in our dataset as hourly based statistics, we plot the volume distribution in Figure 4. Again, we note that trading activities are well reflected by tweet activities. The volume starts to rise drastically two or three hours before the market opens, and then reaches a peak at 9:00pm. It drops during the lunch time and reaches the second peak around 2:00pm (after lunch). Above observations clearly show that market dynamics are discussed in tweets and the content in tweets’ discussion very well reflects the fine- grained aspects of stock markettrading,openingandclosing. Figure 1 . Tweet activity around $aapl’s earnings report date on Jan 23 2013. Figure 2 . Tweet volume distribution in our data over hours averaged across each day. 4. PROPOSED SYSTEM We take news data as an example to introduce the architecture of the neural network model adopted in this work. The model is built for each particularS&P500rm.The goal of the model is to predict a stock price y that is close to the real stock price y of the rm. The raw input of each model is the set of financial news titles collected on the target rm. Intuitively, news content can be useful forfurtherenhancing the sprediction accuracy. Fig Shows the System architecture: Fig 3 .proposed system architecture Registration - Register user with First name, last name, email, password, confirm password, and contact no. etc.
- 3. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 05 | May 2019 www.irjet.net p-ISSN: 2395-0072 © 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 6092 Fig 4. Registration Fig 5 . login 4.1 Stock Prediction This section demonstrates the effectiveness of our SSN based approach for stock prediction. We leverage the sentiment time-series on two kinds of topics from SSN: 1). Node topic from the target stock itself, 2). Neighbor node/edge topics. We note that the price correlation stock network (CSN) also defines neighbor relationships basedon the Pearson's correlation coefficient between pair of past price series. We build a two variables VAR model to predict the movement of a stock’s daily closing price. One variableisthe price time series of the target stock another is the covariate sentiment/price time series We setup two baselines according to the sources of the covariate time series as follows: 1. Covariate price time series from CSN, we try the price time series from the target stock’s closest neighbor which takes the maximum historical price correlation in CSN. 2. With no covariate time series, we try the target stock’s price only based on the univariate autoregression (AR) model. To summarize, we try different covariate sentiment (𝑆(.)) or price (𝑃(.)) time series from SSN or CSN together with the target stock’s Figure 6 . Expected output for Prediction (x-axis is the training window size, y-axis is the prediction accuracy) with different covariate sources. Table 1. Performance comparison of the average and best (in parentheses) prediction accuracies over all training window sizes. 5. MATHEMATICAL MODEL Neural Network 1) Import all necessary libraries (NumPy, skicit-learn, pandas) and the dataset, and define x and y. 2)Initialize the weights as between 0 and 1 3)while Optimal Weight not get 4) Propagating forward through NN 5)Get result for Comparing the real values 6) Then Backward Propagation for update weight 7)if Optimal weight not get repeat from step 3 8)Stop 6. ADVANTAGES Biases, gone forever. Cross-checking results become important. Scheduling tasks. Convenience galore. No Advisors required. 7. APPLICATIONS Market simulation Stock Market Analysis
- 4. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 05 | May 2019 www.irjet.net p-ISSN: 2395-0072 © 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 6093 Stock Market Exchange 8. CONCLUSION This paper proposed to build a stock network from co- occurrences of ticker symbols in tweets. The properties of SSN reveal some close relationships between involved stocks, which provide goodinformationforpredictingstocks based on social sentiment. Experimental resultsshowedthat event embeddings-based document representations are better than discrete events-based methods, and deep convolutional neural network can capture longer-term influence of news event than standard feedforward neural network. REFERENCES [1] Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, no. 7553, pp. 436–444, 2015. [2] M. Liu, J. Shi, Z. Li, C. Li, J. Zhu, and S. Liu, “Towards better analysis of deep convolutional neural networks,” IEEE Transactions on Visualization and Computer Graphics, vol. 23, no. 1, pp. 91–100, 2017. [3] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” arXiv:1512.03385, 2015. [4] K. Simonyan, A. Vedaldi, and A. Zisserman, “Deep inside convolutional networks: Visualisingimageclassification models and saliency maps,” arXiv:1312.6034, 2013. [5] L. M. Zintgraf, T. S. Cohen, and M. Welling, “A new method to visualize deep neural networks,” arXiv:1603.02518, 2016. [6] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imagenet classification with deep convolutional neural networks,” in NIPS’12, 2012, pp. 1097– 1105. [7] M. D. Zeiler and R. Fergus, “Visualizing and understanding convolutional networks,” in ECCV’14, 2014, pp. 818–833. [8] D. E. Rumelhart, G. E. Hinton, and R. J. Williams, “Learning representations by back-propagating errors,” Nature, vol. 323, no. 6088, pp. 533–536, 1986. [9] G. Neubig and Others, “DyNet: The dynamic neural network toolkit,” arXiv preprint arXiv:1701.03980, 2017. [10] “Deep learning for event-driven stock prediction,” in IJCAI’15, 2015, pp. 2327–2333. K.