Question 1

This question is related to a hypothetical personalised news reader system which maintains a profile of each user including:

gender, age, job, list-of-interests,

where list-of-interests is a subset of

{Art, Education, Finance, Politics, Science, Society, Sport, Travel}.

Consider the following stereotypes:

PROFESSIONAL_FINANCIAL_READER:

Trigger: (age>22) and (age<60) and (job in Professional-jobs) and ((list-of-interests includes Finance) or (list-of-interests

includes Politics) or (list-of-interests includes Society))

P=0.7

YOUNG_TRAVELLER:

Trigger: (age>16) and age<30) and (list-of-interests includes Travel)

P=0.8

POLITICALLY_ENGAGED_ADULT:

Trigger: (age>25) and (age<60) and ((list-of-interests includes Politics) or (list-of-interests includes Society))

P=0.9

(a) What do the values of the ratings for the facet Business news mean in the stereotypes PROFESSIONAL_FINANCIAL_READER and YOUNG_TRAVELLER? [2 marks]

(b) Cite an example from the tables that illustrates one problem with stereotypes. [2 marks]

(c) Consider a user U with the following profile:

gender=male, age=40, job=manager,

list-of-interests = {Finance, Politics, Sport, Travel}

Calculate the probability for the user to have high interest in business news and high interest in local news. Show your working. [6 marks]

(d) What is a scrutable user model and what are the claimed benefits? Suggest how a scrutable user model could be implemented for this scenario. [6 marks]

(e) Imagine that you are a consultant with expertise in adaptive information systems. You are asked by the news publisher to suggest one further adaptive feature to     add to this system and what extra information you would need to capture for the user profile. [4 marks] [question 1 total: 20 marks]

Question 2

A city council has a strategy to expand the tourism sector for the city. Part of their    plan is to develop a mobile personalised city guide which can recommend to visitors nearby places of interests, good food and special events when they are visiting the  city.

According to Burke (2002), three key features of recommender algorithms are:

Background data;

Input data;

Algorithm.

(a) Compare and contrast collaborative filtering algorithms with content-based algorithms on these three key features in the case of the mobile personalised   city guide outlined above. You answer should include illustrative examples from the city guide, describing how the three key features of recommender algorithms are addressed by each of these two classes of algorithms. [12 marks]

(b) Discuss one benefit and one limitation for both approaches (collaborative

filtering and content-based filtering) to provide recommendations of nearby pleases of interest. [4 marks]

(c) Assume that the city guide includes a new feature to recommend restaurants to a group of users. Paul, John, Marry, and Jane are registered users of the system and  their prediction values for eight restaurants are given in the following matrix:

Use the Least misery strategy to identify a restaurant to recommend to the group. Show your working. Use the example to point at one advantage of the “Least misery strategy” over the “Average strategy” . [4 marks] [question 2 total: 20 marks]

Question 3

This question relates to a hypothetical regional newspaper which covers a wide    range of topics of local interests. The publisher plans to launch a personalised online service on a subscription basis. Readers can subscribe to a list of chosen topics,      which allows them to access articles and announcements that are related to these    topics. This regional newspaper plans to deliver an adaptive feature which will           recommend new articles and announcements that are of interest to the readers.

(a) Use Jameson’s schema, which was introduced in lectures, to explain how user subscription information can be used to offer personalised recommendation of   new articles and announcements. [6 marks]

(b) Consider an adaptive system that recommends articles to users. The matrix M below indicates which articles a1 , …,a7 were liked by which reader r1, …, r5            (value 1 indicates that the reader liked the article). Assume that r3 would like a  suggestion for which articles may be interesting to read. Apply Amazon’s item- item collaborative filtering algorithm to find the most suitable article to            recommend to r3 . Show your working.