L18-21_finalnew [Compatibility Mode].pdf

download L18-21_finalnew [Compatibility Mode].pdf

of 13

Transcript of L18-21_finalnew [Compatibility Mode].pdf

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    1/29

    BITS PilaniPilani Campus

    BITS Pilani

    presentationN.MEHALA

    FACULTY,CS/IS GROUP

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    2/29

    BITS Pilani, Deemed to be University under Section 3 of UGC Act, 1956

    INFORMATION RETRIEVAL

    4/14/2015 2CS F469

    CS F469Second Semester 2014-15

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    3/29

    BITS Pilani, Pilani Campus

    3

    What is Recommendation systems?

    Items

    Search Recommendations

    Products, web sites, blogs, news items, …

    Recommendation systems areprograms which attempt to predict

    items that a user may be interested inThese systems serve two important

    functions:They help users deal with theinformation overload by giving

    them recommendations ofproducts, etc.They help businesses makemore profits, i.e., selling moreproducts.

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    4/29

    BITS Pilani, Pilani Campus

    4

    Example applications

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    5/29

    BITS Pilani, Pilani Campus5

    What book should I buy?

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    6/29

    BITS Pilani, Pilani Campus6

    What news should I read?

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    7/29

    BITS Pilani, Pilani Campus

    The Recommendation Problem• We have a set of users U and a set of items S to

    be recommended to the users.

    • Let p be an utility function that measures the

    usefulness of item s (∈∈∈∈ S ) to user u (∈∈∈∈ U ), i.e.,

     – p :U ×

    S→

    R , where R is a totally ordered set(e.g., non-negative integers or real numbers ina range)

    • Objective – Learn p based on the past data

     – Use p to predict the utility value of each item s 

    (∈ S ) to each user u (∈ U ) 7

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    8/29

    BITS Pilani, Pilani Campus

    As Prediction

    • Rating prediction, i.e., predict the rating score

    that a user is likely to give to an item that s/hehas not seen or used before. E.g.,

     – rating on an unseen movie. In this case, the

    utility of item s to user u is the rating given to s by u .

    • Item prediction, i.e., predict a ranked list of items

    that a user is likely to buy or use.

    8

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    9/29

    BITS Pilani, Pilani Campus

    Two basic approaches

    • Content-based recommendations:

     – The user will be recommended items similar tothe ones the user preferred in the past

    • Collaborative filtering (or collaborative

    recommendations):

     – The user will be recommended items thatpeople with similar tastes and preferences liked

    in the past.

    • Hybrids: Combine collaborative and content-based methods.

    9

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    10/29

    BITS Pilani, Pilani Campus

    10

    Content-BasedRecommender System

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    11/29

    BITS Pilani, Pilani Campus

    11

    Plan of action

    likeslikes

    Item profilesItem profiles

    RedRedCirclesCirclesTrianglesTriangles

    User profileUser profile

    matchmatch

    recommendrecommend buildbuild

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    12/29

    BITS Pilani, Pilani Campus12

    Content-Based Recommendation• In content-based  recommendations the system

    tries to recommend items that matches the UserProfile.

    • The Profile is based on items user has liked in the

    past or explicit interests that s/he defines.• A content-based recommender system matches

    the profile of the item to the user profile to decide

    on its relevancy to the user.

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    13/29

    BITS Pilani, Pilani Campus13

    Simple Example

    Read update User Profile

    New books User Profile

    RecommenderSystems

    Match

    recommendation

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    14/29

    BITS Pilani, Pilani Campus

    Advantages

    • No need for data on other users

     – No cold-start or sparsity problem

    • Able to recommend to users with unique tastes

    • Able to recommend new and unpopular item

     – No first-rater problem

    14

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    15/29

    BITS Pilani, Pilani Campus

    15

    Limitations of content-based approach

    • Finding the appropriate features

     – e.g., images, movies, music• Overspecialization

     – Never recommends items outside user’s

    content profile – People might have multiple interests

    • Recommendations for new users

     – How to build a profile? – A new user, having very few ratings, would

    not be able to get accurate recommendations.

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    16/29

    BITS Pilani, Pilani Campus

    16

    Collaborative Filtering

    • Consider user c

    • Find set D of other users whose ratings are“similar” to c’s ratings

    • Estimate user’s ratings based on ratings of

    users in D

    Set of other usersSimilar

    RatingsRatings

    Estimate

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    17/29

    BITS Pilani, Pilani Campus

    Limitation of Collaborative method• User’s rating problem

     – Different users might use different scales• Sparsity (Long Tail problem)

     – The number of ratings already obtained is usually very smallcompared to the number of ratings that need to be predicted

    • Scalability – System typically have to search millions of users and items, it causes

    a serious scalability problem

     – However, these correlations will change when new users are added• Adaptability – Requirement of a user may change over time

    • New User Problem – Must first learn the user’s preferences from the ratings that the user

    gives• New Item Problem

     – Until the new item is rated by a substantial number of users, therecommender system would not be able to recommend it

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    18/29

    BITS Pilani, Pilani Campus

    Hybrid Methods

    • Content-based and collaborative methods have

    complementary strengths and weaknesses• Combine methods to obtain the best of both

    • Various hybrid approaches:

     – Apply both methods and combine recommendations

     – Use collaborative data as content

     – Use content-based predictor as another collaborator

    18

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    19/29

    BITS Pilani, Pilani Campus

    19

    Search engines andrecommender systems

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    20/29

    BITS Pilani, Pilani Campus

    Search Engines vs.

    Recommender Systems

    Search Engines

    Goal – answer users adhoc queries

    Input – user ad-hoc needdefined as a query

    Output- ranked itemsrelevant to user need(based on user

    preferences???)Methods - Mainly IRbased methods

    Recommender Systems

    Goal – recommend servicesor items to user

    Input - user preferencesdefined as a profile

    Output - ranked items basedon user preferences

    Methods – variety ofmethods, IR, ML

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    21/29

    BITS Pilani, Pilani Campus

    Query Recommender

    • Content based or content ignorant or hybrid

    recommender for queries using informationabout documents

    21

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    22/29

    BITS Pilani, Pilani Campus

    Content-Boosted Collaborative Filtering

    22

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    23/29

    BITS Pilani, Pilani Campus14.04.2015

    Context-Based Recommender Systems

    23

    The recommender system uses additional data about

    the context of an item consumption.For example, in the case of a restaurant the time or

    the location may be used to improve therecommendation compared to what could be

    performed without this additional source ofinformation.

    A restaurant recommendation for family celebration

    should be different than a restaurant recommendationfor official party.

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    24/29

    BITS Pilani, Pilani Campus14.04.2015

    Recommend a vacation

    Winter vs. summer

    Recommend a purchase (e-retailer)

    Gift vs. for yourself 

    Context-Based Recommender Systems:Motivating Examples

    24

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    25/29

    BITS Pilani, Pilani Campus

    What is the user when asking for a

    recommendation?Where (and when) the user is ?

    What does the user (e.g., improve hisknowledge or really buy a product)?

    Is the user or with other ?

    Are there products to choose or only ?

    25

    Context-Based Recommender Systems

    What simple recommendation techniques ignore?

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    26/29

    BITS Pilani, Pilani Campus

    Context-Based Recommender Systems

    Major obstacle for contextual computing

    Obtain sufficient and reliable data describing the user context

    Selecting the right information, i.e., relevant in a particular

    personalization task

    Understand the impact of contextual dimensions on the

    personalization process

    Computational model the contextual dimension in a more

    classical recommendation technology

    For instance: how to extend Collaborative Filtering to include

    contextual dimensions?

    26

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    27/29

    BITS Pilani, Pilani Campus

    27

    Evaluating Recommendations• Precision

     – Accuracy of predictions

     – Compare predictions with known ratings

    • Recommendation Quality

     – Top-n measures (e.g., Breese score)

    • Item-Set Coverage – Number of items/users for which system can make

    predictions

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    28/29

    BITS Pilani, Pilani Campus

    Evaluating Recommendations

    28

     

     

     

    Receiver operating characteristic (ROC)• Tradeoff curve between false positives and falsenegatives

    Assuming the confusion matrix above,its corresponding table of confusion,for the cat class, would be:

  • 8/9/2019 L18-21_finalnew [Compatibility Mode].pdf

    29/29

    BITS Pilani, Pilani Campus

    29

    The false positive rate is FPR= .Where FP is number of false positives, and TN is number of true negatives.

    True positive rate (or sensitivity): TPR =TP  /(TP +FN )

    It is a plot of the true positive rate against the false positive rate forthe different possible cutpoints of a test.

    Receiver operating characteristic (ROC)