TE4-2 30th Fuzzy System Symposium(Kochi,September 1-3,2014) Advertising Slogan Selection System Using Review Comments on the Web. 2 Masafumi Hagiwara

Similar documents
Topic Models and Applications to Short Documents

Your Virtual Workforce. On Demand. Worldwide. COMPANY PRESENTATION. clickworker GmbH 2017

Programming in Japanese for Literacy Education

Automated Slogan Production Using a Genetic Algorithm

USING TOPONYM CO-OCCURRENCES TO MEASURERELATIONSHIPS BETWEEN PLACES

CORE: Context-Aware Open Relation Extraction with Factorization Machines. Fabio Petroni

Replicated Softmax: an Undirected Topic Model. Stephen Turner

Applied Natural Language Processing

December 3, Dipartimento di Informatica, Università di Torino. Felicittà. Visualizing and Estimating Happiness in

Utilizing Portion of Patent Families with No Parallel Sentences Extracted in Estimating Translation of Technical Terms

Reductionist View: A Priori Algorithm and Vector-Space Text Retrieval. Sargur Srihari University at Buffalo The State University of New York

EVERYTHING YOU NEED TO KNOW ABOUT THE SCORPIO ZODIAC SIGN - ASTROLOGY, COMPATIBILITY, LOVE, TRAITS AND PERSONALITY (EVERYTHING YOU NEED TO

Large-Scale Matrix Factorization with Distributed Stochastic Gradient Descent

Computational Oracle Inequalities for Large Scale Model Selection Problems

B u i l d i n g a n d E x p l o r i n g

Multi-theme Sentiment Analysis using Quantified Contextual

Language as a Stochastic Process

Boolean and Vector Space Retrieval Models CS 290N Some of slides from R. Mooney (UTexas), J. Ghosh (UT ECE), D. Lee (USTHK).

Multimedia analysis and retrieval

A function is a rule that establishes a relationship between two quantities, called

Data Mining Recitation Notes Week 3

An Empirical Study on Dimensionality Optimization in Text Mining for Linguistic Knowledge Acquisition

Structured Neural Networks (I)

Neural Turing Machine. Author: Alex Graves, Greg Wayne, Ivo Danihelka Presented By: Tinghui Wang (Steve)

Lesson 5: Solving Equations

NMR SPECTROSCOPY IN INORGANIC CHEMISTRY (OXFORD CHEMISTRY PRIMERS) BY JONATHAN A. IGGO

Linear Classifiers IV

Algorithmic Methods of Data Mining, Fall 2005, Course overview 1. Course overview

Aspect Term Extraction with History Attention and Selective Transformation 1

The Curse Of Rocky Colavito: A Loving Look At A Thirty-Year Slump By Terry Pluto

ALGEBRA II SEMESTER EXAMS PRACTICE MATERIALS SEMESTER (1.2-1) What is the inverse of f ( x) 2x 9? (A) (B) x x (C) (D) 2. (1.

GEOCHEMISTRY OF NATURAL WATERS BY JAMES DREVER DOWNLOAD EBOOK : GEOCHEMISTRY OF NATURAL WATERS BY JAMES DREVER PDF

a) b) (Natural Language Processing; NLP) (Deep Learning) Bag of words White House RGB [1] IBM

SYMBOLIC INTERACTIONISM: AN INTRODUCTION, AN INTERPRETATION, AN INTEGRATION: 10TH (TENTH) EDITION BY JOEL M. CHARON

Multi-faceted Learning for Web Taxonomies

Probabilistic Context Free Grammars. Many slides from Michael Collins

THE MOST IMPORTANT BIT

Exploring Urban Areas of Interest. Yingjie Hu and Sathya Prasad

Navigating to Success: Finding Your Way Through the Challenges of Map Digitization

Finding Main Streets: Applying Machine Learning to Urban Design Planning

Machine Learning for Physicists Lecture 1

Leverage Sparse Information in Predictive Modeling

Summarizing Opinions: Aspect Extraction Meets Sentiment Prediction and They Are Both Weakly Supervised

Neural networks CMSC 723 / LING 723 / INST 725 MARINE CARPUAT. Slides credit: Graham Neubig

Bio-Medical Text Mining with Machine Learning

Enabling ENVI. ArcGIS for Server

Boolean and Vector Space Retrieval Models

Chinese Character Handwriting Generation in TensorFlow

Better Decisions for Your Business Drive your company forward with ultra-precise weather forecasts

Machine Learning. Boris

Nonlinear Characterization of Activity Dynamics in Online Collaboration Websites

Natural Language Processing and Recurrent Neural Networks

Metadata Catalogue of Diatom Names

PhysicsAndMathsTutor.com. International Advanced Level Statistics S2 Advanced/Advanced Subsidiary

EVOLUTION OF PHYSICS BY ALBERT EINSTEIN AND LEOPOLD INFELD DOWNLOAD EBOOK : EVOLUTION OF PHYSICS BY ALBERT EINSTEIN AND LEOPOLD INFELD PDF

BY DEBORAH BARRETT - LEADERSHIP COMMUNICATION (4TH EDITION) ( ) [HARDCOVER] BY DEBORAH BARRETT

Bayesian Contextual Multi-armed Bandits

Deep Learning for Natural Language Processing

Advanced/Advanced Subsidiary. You must have: Mathematical Formulae and Statistical Tables (Blue)

Full file at

Hidden Markov Models. x 1 x 2 x 3 x N

Recent Advances in Bayesian Inference Techniques

ECS 120: Theory of Computation UC Davis Phillip Rogaway February 16, Midterm Exam

CIKM 18, October 22-26, 2018, Torino, Italy

4452 Mathematical Modeling Lecture 16: Markov Processes

Deep Sequence Models. Context Representation, Regularization, and Application to Language. Adji Bousso Dieng

Fertilization of Case Frame Dictionary for Robust Japanese Case Analysis

IE598 Big Data Optimization Introduction

BIOINFORMATICS: METHODS AND APPLICATIONS: (Genomics, Proteomics and Drug Discovery)

NSIDC Metrics Report. Lisa Booker February 9, 2012

15 Introduction to Data Mining

LASER TEAM ANTI-BULLYING PROGRAM STUDENT WORKSHEET MASTERS AND ANSWER KEYS

Knowledge Discovery. Zbigniew W. Ras. Polish Academy of Sciences, Dept. of Comp. Science, Warsaw, Poland

CHAPTER 2: DATA MINING - A MODERN TOOL FOR ANALYSIS. Due to elements of uncertainty many problems in this world appear to be

Attention Based Joint Model with Negative Sampling for New Slot Values Recognition. By: Mulan Hou

Text Mining. Dr. Yanjun Li. Associate Professor. Department of Computer and Information Sciences Fordham University

Recurrent Neural Networks. Jian Tang

Markov Logic Networks for Spoken Language Interpretation

B.Tech (Electronics & Computer Engineering)

Reliable and Interpretable Artificial Intelligence

ABRAHAM LINCOLN: A COMPLETE BIOGRAPHY BY LORD CHARNWOOD DOWNLOAD EBOOK : ABRAHAM LINCOLN: A COMPLETE BIOGRAPHY BY LORD CHARNWOOD PDF

Key Questions and Issues. What is GIS? GIS is to geographic analysis as: What is GIS? 9/3/2013. GEO 327G/386G, UT Austin 1

A Study for Evaluating the Importance of Various Parts of Speech (POS) for Information Retrieval (IR)

Google Adwords. 8WEB Google Adwords. Capture leads & make sales. Y o u r P a r t n e r s I n O n l i n e S a l e s

BACHELOR OF TECHNOLOGY DEGREE PROGRAM IN COMPUTER SCIENCE AND ENGINEERING B.TECH (COMPUTER SCIENCE AND ENGINEERING) Program,

Cost and Preference in Recommender Systems Junhua Chen LESS IS MORE

Portals: Standards in Action

Association Rule Mining on Web

A Little History of Machine Learning

A directory of information resources on radioactive waste management, decontamination and decommissioning, and environmental restoration

What is this Page Known for? Computing Web Page Reputations. Outline

Social media and the news industry

The Ultimate Guide To Chatbots For Businesses ONLIM 2018

BAYESIAN MODELS FOR ASTROPHYSICAL DATA: USING R, JAGS, PYTHON, AND STAN BY JOSEPH M. HILBE, RAFAEL S. DE SOUZA, EMILLE E. O.

Solving Inequalities Using Addition or Subtraction 7.6. ACTIVITY: Writing an Inequality. ACTIVITY: Writing an Inequality

(i) The optimisation problem solved is 1 min

Marine Biology, 6th Edition By Peter Castro, Michael E. Huber

1 Handling of Continuous Attributes in C4.5. Algorithm

CSCI 315: Artificial Intelligence through Deep Learning

Syllabus Structure for Computer Science and Systems Engineering

Transcription:

Web Advertising Slogan Selection System Using Review Comments on the Web 1 1 Hiroaki Yamane, 2 2 Masafumi Hagiwara 1 1 Graduate School of Science and Technology, Keio University Abstract: Increased demand for web advertising has resulted in a corresponding increase in the need to develop personalized advertisements targeted at individuals online. We propose an automated advertising slogan selection system that can satisfy this requirement. Many customer reviews and comments are available publicly on online shopping sites. The proposed system uses content mining to extract favorable reports from the web and arranges the data into a specific knowledge representation structure to improve the advantage of the target product. For a particular business, the proposed system first extracts tuples, composed of elements that express the knowledge representation from each user-written review. Then, these tuples are selected using a frequency-based approach and emotion corpus. Subsequently, for each tuple, advertising slogans are chosen from the advertising slogan corpora using a neural network. For verification, we used data from an electronic commerce website for hotels to evaluate two aspects of our system (namely, quality of selected tuples and advertising slogans). The results of the experiments confirm that the proposed system can extract suitable tuples when the given data are sufficient. It can also retrieve slogans even when their meanings are convoluted. 1 EC 2012 Web 1 [1] EC EC EC EC Web (1) (2) (1) (2) 2 3 4 2 1 EC (: ) (: ) (:,,) 654

EC Taisetsuna ano hito to, kitto nanndomo otozureru machi ni naru. Without doubt, you will return many times to this lovely town. 2008/Yomiuri(the Japanese Newspaper Publisher)/the Austrian Airlines/ You can fully enjoy fascinating Vienna in-depth Main Slogan Metadata 2: (=) 1: 2.1 EC [2] 2.2 [3] T j j F t () j n n j=1 F t(t j ) n j=1 F a(t j ) S t S t = ( ) Ft (T j ) n F F j=1 t(t j ) t (T j ) ( ) (1) Fa (T j ) n F j=1 a(t j ) F t (T j ) [4] 8 2.3 2 655

w j II k II w i II j II w j I k I x i I j I S a S S 1 2 3 4 n Other Slogans Selection Other slogans Pre-constructed Words in slogan Slogan Selection Slogans Metadata Link Metadata Words in Metadata Sentences containing extracted 3-tuples 3: [5] [6] 3 (1) (2) t ji F ki (t ji ) t ji F Metadata (t ji ) w ji k i w ji k I = F ki (t ji ) F Metadata (t ji ) (2) t jii F iii (t jii ) t jii F Slogan (t jii ) w iii j II w iii j II = F i II (t jii ) F Slogan (t jii ) (3) Slogans Other slogans 2.3.1 ( Slogans ) i F ii (t ji ) t ji X(i I ) = {x ii 1, x ii 2,..., x ii N ji } x ii j I = F ii (t ji ) (4) X(i I ) w ji k I k S S (i I, k I ) S S (i I, k I ) = N ji j I =1 w ji k I x ii j I = N ji j I =1 F ii (t ji )F ki (t ji ) F Metadata (t ji ) S S (i I, k I ) ( Slogan ) (5) 656

2.3.2 ( Other slogans ) S S (i I, k I ) F kiii t jii F kiii (t all ) ( ) ( Other slogans ) w jii k II 1: (1-5) (1-5) (1-5) (s) 4.23 3.57 3.91 10.73 3.81 3.30 3.66 9.95 w jii k II = F ki II (t jii ) F kiii (t all ) (6) F kiii (t all ) S a (S s, k II ) S a (S s, k II ) = = N jii j II =1 N jii j II =1 w iii j II w jii k II S S (7) F iii (t jii )F kiii (t jii ) F Slogan (t jii )F kiii (t all ) S S (8) S a (S s, k II ) (1) (2) 3 3.1 3.1.1 [7] 100 ( 85,052 ) KNP [8] [9] 20 (1) 20 40 8 2: 1 2 (5) (5) 2,000 24,439 () (8) 2,000 (8) 500 (5) (8) () 2,000 3.1.2 (i) ( 5:-1:) (ii) ( 5:-1:) (iii) ( 5:-1:) 3.1.3 1 3.2 2 1 2 1 ( Slogans ) ( Other slogans ) 2 Other slogans (5) (8) (5) (8) 657

3: 1 (1-5) (1-5) (1-5) (s) 3.34 3.87 3.73 18.4 2.75 3.71 3.34 13.9 4: 2 (1-5) (1-5) (1-5) (s) 2.86 3.33 3.34 16.3 2.68 3.51 3.52 18.5 2.99 3.49 3.50 17.7 3.2.1 4 [10, 11, 12, 13] 24,439 1 500 500 [14] 8 2 2 10 40 2 [15] 2,000 9 3 2 20 120 1 2 3.2.2 (i) ( 5:-1: ) (ii) ( 5:-1:) (iii) ( 5:-1: ) 3.2.3 3 1 4 2 1 2 4 Web bag-of-words 5 245423. [1] Forbes: Online ad spending tops $100 billion in 2012, http://www.forbes.com/sites/roberthof/ 2013/01/09/online-ad-spending-tops-100- billion-in-2012/, 2013. [2], :,, Vol.49, No.7, pp.2598-2603, 2008. 658

[3] Hiroaki Yamane, Masafumi Hagiwara: Tag line generating system using knowledge extracted from statistical analyses, AI & Society 28 pp.1-11, 2013. [4] :,, 1993. [5] J. L. Elman: Distributed representations, simple recurrent network, and grammatical structure, pp. 195-225 Machine Learning 1991. [6] Tsukasa Sagara, Masafumi Hagiwara: Natural language neural network and its application to question-answering systems, The International Joint Conference on Neural Networks, pp.1367-1373, 2012. [7] : http://travel.rakuten.co.jp/group/ tiku/ [8] : KNP, http://nlp.ist.i.kyotou.ac.jp/index.php?knp [9], :,, Vol.15, No.2, pp.101-136, 2008. [10] :,, 1998. [11] : 2,, 2005. [12] :,, 2008. [13] : 3,, 2011. [14],,,,,,,,, : 500,, 2011. [15] Make1:, http://catch copy.make1.jp/index.cgi 223-8522 3-14-1 yamane@soft.ics.keio.ac.jp hagiwara@soft.ics.keio.ac.jp 659