Motivation and Overview Towards Defining and Exploiting 12.3.40.65 - - PDF document

1

Towards Defining and Exploiting Similarities in Web Application Use Cases through User Session Analysis

Sreedevi Sampath, University of Delaware Amie Souter, Drexel University Lori Pollock, University of Delaware

Workshop on Dynamic Analysis (WODA), May 25, 2004 co-located with International Conference on Software Engineering (ICSE 2004)

Motivation and Overview

12.3.40.65 GET index.jsp 12.3.40.65 GET login.jsp 12.3.40.65 GET reg.jsp? login=xxx&password=hello& 12.3.40.65 GET myinfo.jsp

View user sessions as use cases Behaviorally related sequence of events performed by the user through interaction with the system User session analysis Test case generation Monitoring load of traffic Content personalization Test suite reduction Software development/ maintenance tools

• Clustering via concept analysis
• Common subsequences analysis

Learn about dynamic behavior

Step 1 Clustering via Concept Analysis

• Mathematical technique for clustering objects

that have common discrete attributes

• Set of objects, O: user sessions, us
• Set of attributes, A: URLs, u
• Relation, R: us requests u
• Concept analysis identifies all concepts (Oi, Aj)

for a given tuple (O, A, R)

x x GB x x x x x GL x x PL x x x us4 x x us5 x x x us6 x x x us3 x x us1 GM GS GR GD us2 x x x x x

,{GD,GR,GL})

({us1, us2, us3, us4, us5, us6}

,{GD,GR,GL,GS}) ,{GD,GR,GL,PL,GS}) ,{GD,GR,GL,GS,GB})

({us3} ({us4} ({us2} ,{GD,GR,GL,GS,GM}) ({us6} ({us2, us3, us4, us6}

,{GD,GR,GL,PL,GS,GB}) {GD,GR,GL,PL,GS,GB,GM})

(null, CONCEPT LATTICE: FULL

• b

j e c t s attributes (URLs) RELATION TABLE GD GR GL us1 us5 GS PL GB us3 us2 GM us6 SPARSE us4

Step 2 Heuristic for Test Suite Reduction

• Smallest set of user sessions
• Covers all the URLs
• Represents common URL subsequences
• f different use cases

us1 us5 GS PL GB us3 us4 us2 GM us6 GD GR GL Resulting reduced test suite: {us2, us6} Identify next-to-bottom nodes Pick one user session from each of these next-to-bottom nodes

Hypothesis Motivating the Approach

• Common Subsequences Hypothesis:

The set of user sessions clustered together into the same concept node will have a high commonality in the subsequences of URLs in their sessions

2

Finding Common Subsequences of URLs

NODE 003

• bjects

{ us3, us6 } attributes { GD, GL, GR, GS, PL } us3 GD GR GL PL GS PL GR GL PL GS us6 GD GR GL GB PL GS GR GL GB PL GS Common Subsequences [GD, GR, GL] [PL, GS] [GR, GL]

Subsequences of URLs are representative of partial use cases of the user sessions

Metric for Common Subsequences Hypothesis

• attr-size[n] set: level of node in lattice

# user sessions= 4 # URLs = 5 # user sessions = 3 # URLs = 5 attr-size[5]: level 5 {a, b, c, d, e} 80% abc, abe 3 80% ab, bc, be 2 100 % a, b, c, d, e 1 Percent attrs covered Common subsequence Sub seq size us1 us2 us3 a b c d a b e a b c e a b e d a b c d a b e a Metric Percent of attributes covered by common subsequences

• f URLs of various sizes

Experiment: Applications Used

• Bookstore web application

9,748 LOC, 385 methods, 11 classes Front end: JSP, Backend: MySql 123 user sessions

• uPortal application

38,589 LOC, 4233 methods, 508 classes Java, JSP, XML, J2EE 2083 user sessions

Results for Common Subsequences Hypothesis

Bookstore

1 4 710 13161922252831343740

Attr-Size Set

8

Subsequence Size

24 26 28 30 32 34 36 38 40 42 44 46 48 50 52

Percent of Attributes Covered

Results for Common Subsequence Hypothesis

Bookstore Result: subsequences of various sizes cover reasonable percent of attributes

Conclusions for Common Subsequences Hypothesis

• Between user sessions of a node there

exists commonality in subsequences of URLs

• These common subsequences cover a

reasonable percent of URLs (attributes)

• f the node
• Clustering based on single URLs

clusters similar use cases can choose one object from each node

3

Next-to-bottom Coverage of Use Cases Hypothesis

reduced set {us2, us6} remaining set {us1, us3, us4 us5}

user sessions belonging to next-to-bottom nodes all other user sessions except sessions belonging to next-to-bottom nodes

GD GR GL us1 us5 GS PL GB us3 us4 us2 GM us6

In addition to covering all the URLs of the original test suite, the user sessions in next-to-bottom nodes execute a high percentage of the subsequences of URLs of the rest

• f the original test suite

Results for Next-to-bottom Coverage

• f Use Cases Hypothesis

Bookstore

Result: short sequences present but long sequences are missing Metric: loss of coverage of use cases in remaining set by the reduced set

Conclusion for Next-to-bottom Coverage of Use Cases Hypothesis

• Long sequences absent but smaller

sequences are present in reduced set

• reduced set contains more URLs hence

may contain other URL sequences absent in remaining set

• Moderately supports picking next-to-

bottom nodes for reduced test suite

Pros and Cons of Our Approach

+ Results from common subsequences hypothesis support using concept analysis for clustering user sessions + Experiments show little coverage loss (tech report) by reduced test suite

• Results from next-to-bottom coverage of use

cases hypothesis indicate further work needed on heuristic

Future Work

• Explore additional heuristics
• Additional user session analysis

Useful for other software engineering tasks