Hidekazu Yanagimoto

Sentiment Analysis of Stock Market News with Semi-supervised Learning

In these days, there are many news on stock market on the Internet and investors have to understand them immediately to invest in a stock market. In this study we determine sentimental polarities of the stock market news using a polarity dictionary, which consists of terms and their polarities. To achieve our aim we have to construct the polarity dictionary automatically because of decrease of human efforts. In construction the dictionary we use a semi-supervised learning approach. In the semi-supervised approach at first we make a small polarity dictionary, which a word polarity is determined manually, and using many stock market news, which polarities are not known, new words are added in the polarity dictionary. In this paper we proposed an automatically dictionary construction approach and sentiment analysis of stock market news using the dictionary. To discuss our proposed method we compare polarities determined by a financial expert with polarities determined with our proposed method. Hence, we confirm that the proposed method can make an appropriate dictionary.

Web news classification using neural networks based on PCA

In this paper, we propose a news web page classification method (WPCM). The WPCM uses a neural network with inputs obtained by both the principal components and class profile-based features (CPBF). The fixed number of regular words from each class will be used as a feature vectors with the reduced features from the PCA. These feature vectors are then used as the input to the neural networks for classification. The experimental evaluation demonstrates that the WPCM provides acceptable classification accuracy with the sports news datasets.

Relationship strength estimation for social media using Folksonomy and network analysis

We propose a relationship strength estimation method in social media. We estimate relationship strength between web pages in social bookmarking services using a tag vocabulary and construct a network of the web pages. In this step Bayes theorem is used to estimate true strength from each users strength estimation. After estimation using tags the network is represented in lower dimension space and some non-important links are removed. In this step the network is approximated keeping neighborhood of data in the original network. To evaluate our proposed method we carry out some experiments using artificially generated data and real social bookmarking data. And we confirm that 1) the proposed method can estimate more appropriate relationship strength than ordinary methods based on cooccurrence frequency and tags sharing rate and 2) the proposed method can remain essential links and delete pseudo relationship.

Italian Lira classification by LVQ

In this paper, a new method to classify the Italian Liras by using the learning vector quantization (LVQ) is proposed. The Italian Liras of 8 kinds, 1000, 2000, 5000, 10000, 50000 (new), 50000 (old), 100000 (new), 100000 (old) Liras with four directions A,B,C, and D are used, where A and B mean the normal direction and the upside down direction and C and D mean the reverse version of A and B. The original image with 128 by 64 pixels is observed at the transaction machine in which rotation and shift are included. After correction of these effects, we select a suitable area which shows the bill image and feed the image with 64 by 15 pixels to a neural network. Although the neural network of the LVQ type can process in any order of the dimension of the input data, the smaller size is better to achieve a faster convergence.

User Intent Estimation from Access logs with Topic Model

Abstract As the Internet is widespread and there are many online shops in the Internet, many persons buy products in the online shops. Customers behavior in the online shops is a sequence of customer driven activities intrinsically because his/her movement in an online shop occurs according to only his/her decision. Hence, to achieve satisfactory purchase experiments it is important how the shop supports them. Online shops have to predict visitors’ intents correctly to support them effectively. One of information resources the shops can use is an access log including information on customers movement in the online shop. If they are histories of customers behaviors in online shops and the behaviors depend on customers intents, we can extract new knowledge on them from the access logs. Speaking concretely, we can predict customers’ intents from the access logs since their internal intents affect their activities. We can realized more appropriate recommendation service by changing recommendation strategy depending on customers intents. In this paper, we propose a method to predict customers intents from access logs in a real online shop. We adopt a Topic Tracking Model (TTM) to analyze the access logs.

Recommendation from access logs with ensemble learning

Many recommendation systems find similar users based on a profile of a target user and recommend products that he/she may be interested in. The profile is constructed with his/her purchase histories. However, histories of new customers are not stored and it is difficult to recommend products to them in the same fashion. The problem is called a cold start problem. We propose a recommendation method using access logs instead of purchase histories, because the access logs are gathered more easily than purchase histories and the access logs include much information on their interests. In this study, we construct user’s profiles using product categories browsed by them from their access logs and predict products with Gradient Boosting Decision Tree. In addition, we carry out evaluation experiments using access logs in a real online shop and discuss performance of our proposed method comparing with conventional machine learning and Support Vector Machine (SVM). We confirmed that the proposed method achieved higher precision than SVM over 10 data sets. Especially, under unbalanced data sets, the proposed method is superior to SVM.

Customer state estimation with Poisson distribution model

In this paper, a new access log analysis is proposed which estimates both active states and inactive states from observations simultaneously. I improved burst analysis and developed enthusiasm analysis to detect not only active states but also inactive states. Speaking concretely, I constructed a generative model that assumed observations, which meant event occurrence frequency in this paper, were generated under some Poisson distributions. The generative model based on Poisson distributions consists of some distributions including

Information filtering using a probabilistic model

\lambda