e-business intelligence and relationships

e-business intelligence and relationships

1. Exercises

   • 1. RapidMiner
   • 2. Data pre-processing
   • 3. Prediction models
   • 4. Decision trees
   • 6. Support vector machines
   • 5. Artificial neural networks
   • 7. Model evaluation
   • 8. Association rules

2. Lectures

   • Basics
     • Relationship management
       • value chain/network CRM=get customers and build LT relationships win win add value. Traditional CRM 1.0: sales, mktg, service&support VS Social CRM 2.0. real time interaction e.g. dialog 2 way company/customer Components CRM: Analytical CRM= ERP, DW, Data mining Customer insights  Operational CRM = Content mgmt ,operational DB, Customer touch points...
     • Business intelligence
       • New culture of decision making! Let data drive decisions, not higher paid person opinion HiPPO (before vs now) Data science platforms leaders e.g. IBM, RapidMiner, SAS BI=dikp all that turns data into info into knowledge into plans to drive profitable actions. E.g. data warehouse, biz analysis, content mgmt. DM Data Mining=patterns in data, correlations ML Machine Learning=predict future based on historical data, to automate decisions.  ML algorithm centric -> DM data centric ->  BI user centric 
     • Digital marketing
       • = where mktg meets internet and media e.g. phones, video games e.g. click product, exit, then fb ad. Data driven marketing (big data) Predictive marketing (patterns, trends)
     • Big data
       • Big Data=hi vol, vel, var info in need of info processing for enhanced insight, decision making, and process auto. 1944 BD measured in books BD dimensions 5Vs of BD = velocity, veracity, value, variety, volume Challenge  too much data! BD, info overload, relevant data Data mined from our activity, chats, photo, video, sensors, iot Challenge 2 variety structured CRM POS unstructured email, call, blog Dark Data = data gathered by companies but not used for meaningful purposes Data Lake=large repository of data in raw format BD as mktg tool for music audience engagement data filters auto mktg tools but also in politics
     • Data mining
       • DM=patterns from huge data also known as KDD (knowledge discovery in DB) from data to knowledge CRISP-DM ref model = cross industry standard process for DM def phases, tasks, deliverables for DM projects. Cluster analysis = partitioning set of data objects into subsets Regression= one predictive modeling technique to investigate relationship between dependent and independent variable. e.g. why less calls last week.
     • Machine learning
       • ML=where computers learn w/o being taught to. AI = subfield of computer science, solves tasks easy for humans and hard for pc e.g. face recognition AI>ML>DeepLearning Supervised learning (knowledge output, data labelled) vs Unsupervised (no knowledge of output, no labels on data ML use cases  1. classification e.g. spam filtering, content personalization 2. regression e.g. stock mkt prediction, risk mgmt, weather forecast 3. recommendation e.g. job recruiting, content recommendation 4. imputation e.g. missing customer data MLmodel new data (no label) - model - new data with predicted class historical data (label) split into training data to build model with and test data. Find attributes to influence class assignment. Compare predicted vs actual class. MLworkflow 1 historical data 2 model building 2.1. Model eval 2.2 model optimization 3. New data 4 answers Example of ML products = discover weekly, recommended songs NEW AREA OF MO is DATA LEARNING DL especially used in AI e.g. deep face, deep text, targeted ads on fb side, skype translator, hotness.ai
   • BI and Analytical CRM
     • Methods for customer analysis
       • Decision trees
         • Neural nets, decision trees, SVM - Classification Decision Tree classifies objects by prio. Consist of nodes and edges. Nodes: characteristic attributes Edges: attribute value conditions Leaf nodes: class assignment E.g. salary Less 1000? not credit worthy More 1000? residence? Y credit worthy N not credit worthy Process of application : 1. Use training data 2. Use testing data 3. Application data
         • 1. Build decision tree with training data 2. Evaluate decision tree on test data (accuracy) 3. Apply decision tree to new data w/o class description Attribute AGE Characteristic SEX Class CREDIT WORTHY Overfitting = ref to models accurate on train data for models, but too complex so cannot be used in data in modeling set (under- appropriate- over-)
         • Pruning = simplify a DT by removing error prone components and avoid overfitting Pre=during train of DT Post= after tree is created
       • Artificial neural networks
         • Brain 10bn neurons Human brain as role model Artificial neuron inputs outputs
       • Support vector machines SVM
         • SVM set of methods for supervised learning for clsssification and regression E.g. image classification, hand writing character recognition. Decision surface needed. E.g. boats PROS AND CONS OF THE PAST 3
       • Association rules
         • Market basket analysis = id purchase habits by sold products analysis e.g. sell complementRy products! Association rules E.g. Amazon customers who bought this also bought
       • Evaluation
         • ML model validation Goal = how we’ll train model performs on train data Problem = model optimism Solution = validation techniques Accuracy =#correct predictions/#total data points How often the classifier makes the correct prediction
     • Applications for customer analysis
       • Multi-media-mining
         • Sentiment filtering!! Feedback or complaints Chat bots Text mining = extract of relevant correlations and knowledge unstructured text Linguistic analysis Call center conversation Text sources - text storage - statistical and linguistically analysis - applications Audio mining = recognize and analyze audio signal = speech recognition Speech recognition Recommendation audio tag
         • Image mining =focus on extract patterns, implicit knowledge, image data Vs computer vision (just single image) Meta data = e.g. time of pic creation, gps location Context data = time upload, view or share by others Image mining process via comparing the color distributor histograms
         • Video mining Or computer vision is the discovery of patterns in visual content Cameras, ultra sound recordings Processing: key figures, social behavior, customer streams Image mining applications Product analytics, color trend detection, spaccio temporal analysis of image data, data driven services, outfit browser, social CRM profiling
         • Video mining applications Hurried customers Goal oriented customers Analysts Explorers We are leaving the Information Age and entering the recommendation age Recommender systems are software tools and techniques providing suggestions for items to be of use to a user. Matrix factorization - user item vs +attribute
       • Social network analysis
         • Id user connections to target more in one go. Sociogram based on comms frequency Nodes are actors Centrality - position of individual actor Centralization - tot network Types of abode Cs are star (center actor max centrality) and circle structure (actors same centrality min centrality)
         • Knowledge management 2.0 Identify the knowledge actor Among helper solved initiator connector expert observer shaker etc Formal and informal social network Social network open fb LinkedIn closed yammer
       • Social media mining
         • SMM extraction and analysis of user generated data on social media Vs SMAnalytics study media metrics to help biz strategy Socia medial growth since 2008 Irc Wikipedia fb yt twitter Socia media mining for customer insights to prevent abuse, theft, flashmob
         • Metric = measuring system to quantify a trend, dynamic, or characteristic E.g. ex plain phenomena traffic
       • Recommended systems
         • Recommended systems are software tools and techniques providing suggestions for items to be used by user
       • Knowledge based systems
         • Knowledge base = Data mining : generate knowledge Knowledge based systems : process knowledge 2 types of knowledge representation Deductive knowledge rule based Inductive knowledge case based CBR = case base reasoning Solve new problems based on solution of old Example restaurant recommend system CBR cycle Retrieve Reuse Revise Retain
     • Data visualization
       • Data visualization = technique to com data or info by encoding it in visual objects contained in graphics Example charts, pie chart, maps, donut chart (Magic quadrant ability to execute vs completeness of vision