User search intent and the EM algorithm
by Richard Heimann, Nathan Danneman, Pradeepta Mishra, Bater Makhabel
R: Mining Spatial, Text, Web, and Social Media Data
- R: Mining Spatial, Text, Web, and Social Media Data
- Table of Contents
- R: Mining Spatial, Text, Web, and Social Media Data
- R: Mining Spatial, Text, Web, and Social Media Data
- Credits
- Preface
- 1. Module 1
- 1. Warming Up
- 2. Mining Frequent Patterns, Associations, and Correlations
- An overview of associations and patterns
- Market basket analysis
- Hybrid association rules mining
- Mining sequence dataset
- The R implementation
- High-performance algorithms
- Time for action
- Summary
- 3. Classification
- Classification
- Generic decision tree induction
- High-value credit card customers classification using ID3
- Web spam detection using C4.5
- Web key resource page judgment using CART
- Trojan traffic identification method and Bayes classification
- Identify spam e-mail and Naïve Bayes classification
- Rule-based classification of player types in computer games and rule-based classification
- Time for action
- Summary
- 4. Advanced Classification
- 5. Cluster Analysis
- 6. Advanced Cluster Analysis
- Customer categorization analysis of e-commerce and DBSCAN
- Clustering web pages and OPTICS
- Visitor analysis in the browser cache and DENCLUE
- Recommendation system and STING
- Web sentiment analysis and CLIQUE
- Opinion mining and WAVE clustering
- User search intent and the EM algorithm
- Customer purchase data analysis and clustering high-dimensional data
- SNS and clustering graph and network data
- Time for action
- Summary
- 7. Outlier Detection
- Credit card fraud detection and statistical methods
- Activity monitoring – the detection of fraud involving mobile phones and proximity-based methods
- Intrusion detection and density-based methods
- Intrusion detection and clustering-based methods
- Monitoring the performance of the web server and classification-based methods
- Detecting novelty in text, topic detection, and mining contextual outliers
- Collective outliers on spatial data
- Outlier detection in high-dimensional data
- Time for action
- Summary
- 8. Mining Stream, Time-series, and Sequence Data
- 9. Graph Mining and Network Analysis
- 10. Mining Text and Web Data
- A. Algorithms and Data Structures
- 2. Module 2
- 1. Data Manipulation Using In-built R Data
- What is data mining?
- Introduction to the R programming language
- Data type conversion
- Sorting and merging dataframes
- Indexing or subsetting dataframes
- Date and time formatting
- Creating new functions
- Loop concepts - the for loop
- Loop concepts - the repeat loop
- Loop concepts - while conditions
- Apply concepts
- String manipulation
- NA and missing value management
- Missing value imputation techniques
- Summary
- 2. Exploratory Data Analysis with Automobile Data
- Univariate data analysis
- Bivariate analysis
- Multivariate analysis
- Understanding distributions and transformation
- Interpreting distributions
- Variable binning or discretizing continuous data
- Contingency tables, bivariate statistics, and checking for data normality
- Hypothesis testing
- Non-parametric methods
- Summary
- 3. Visualize Diamond Dataset
- 4. Regression with Automobile Data
- 5. Market Basket Analysis with Groceries Data
- 6. Clustering with E-commerce Data
- 7. Building a Retail Recommendation Engine
- 8. Dimensionality Reduction
- 9. Applying Neural Network to Healthcare Data
- 1. Data Manipulation Using In-built R Data
- 3. Module 3
- 1. Going Viral
- 2. Getting Started with R
- 3. Mining Twitter with R
- 4. Potentials and Pitfalls of Social Media Data
- 5. Social Media Mining – Fundamentals
- Key concepts of social media mining
- Good data versus bad data
- Understanding sentiments
- Sentiment polarity – data and classification
- Supervised social media mining – lexicon-based sentiment
- Supervised social media mining – Naive Bayes classifiers
- Unsupervised social media mining – Item Response Theory for text scaling
- Summary
- 6. Social Media Mining – Case Studies
- A. Conclusions and Next Steps
- Bibliography
- Index
The Expectation Maximization (EM) algorithm is a probabilistic-model-based clustering algorithm that depends on the mixture model in which the data is modeled by a mixture of simple models. The parameters related to these models are estimated by Maximum Likelihood Estimation (MLE).
Mixture models assume that the data is the result of the combination of various simple probabilistic distribution functions. Given K distribution functions and the jth distribution with the parameter, 
, 
is the set of of all distributions:
The EM algorithm performs in the following way. In the first step, an initial group of model parameters are selected. The expectation step is the second step that performs the calculation of the probability:
The previous equation represents the probability of each data object belonging to each distribution. Maximization is the third step. With the result of the expectation step, update the estimation of the parameters with the ones that maximize the expected likelihood:
The expectation step and maximization step are performed repeatedly until the output matches the ending condition, that is, the changes of the parameter estimations below a certain threshold.
Please take a look at the R codes file ch_06_em.R from the bundle of R codes for the previously mentioned algorithm. The codes can be tested with the following command:
> source("ch_06_em.R")
Determining the user search intent is an important yet a difficult issue with respect to the sparse data available concerning the searcher and queries.
User intent has a wide range of applications, cluster query refinements, user intention profiles, and web search intent induction. Given Web search engine queries, finding user intention is also a key and requirement.
To determine the user interest and preferences, the clicks' sequence upon the search result can be used as a good base.
Web search personalization is another important application with the user search intention. It is related to user context and intention. With the user intention applied, more effective and efficient information will be provided.
-
No Comment