Intelligent Traceability Presented by: Auguste Lalande What are - - PowerPoint PPT Presentation

intelligent traceability
SMART_READER_LITE
LIVE PREVIEW

Intelligent Traceability Presented by: Auguste Lalande What are - - PowerPoint PPT Presentation

Domain-Contextualized Intelligent Traceability Presented by: Auguste Lalande What are trace links? Software development generates artifacts, such as regulatory documents , and system level requirement documents Trace links connect the


slide-1
SLIDE 1

Domain-Contextualized Intelligent Traceability

Presented by: Auguste Lalande

slide-2
SLIDE 2

What are trace links?

 Software development generates artifacts, such as regulatory documents, and

system level requirement documents

 Trace links connect the sections of these documents that correspond to the

same idea

slide-3
SLIDE 3

Action Units

 Break down natural language requirement into its key parts

The Wayside Segment shall transmit Wayside Status Messages (WSMs) to the Automobile Segment. Action transmit Semantic Group Transmissive Properties Agent: wayside segment Theme: WSM Recipient: automobile segment

slide-4
SLIDE 4

Action Units: How?

 POS parse tree (Stanford Parser)  Dependency set (Stanford Parser)  Syntactic group

 Group of verbs with similar syntactic

properties

 Mapping rules

 Given the verb transmit, the subject

corresponds to the agent property.

 agent = nsubj = Wayside Segment

Dependency set dobj(transmit-5, Messages-8) nsubj(transmit-5, Segment-3) ⋮

slide-5
SLIDE 5

Link Heuristics

 Given two action units, should a link be made?  Match the properties of two action units based on their semantic group

 Different strictness levels of matching: 0 (least strict) to 3 (most strict)  A domain specific ontology is used to allow for hierarchical matching and to

relate equivalent terms.

Affirmative-Inclusive Agent match level 0 Theme match level 2

slide-6
SLIDE 6

Overall System

Vector Space Model

Similar to Latent Semantic Indexing

 Rank generated links

slide-7
SLIDE 7

Mean Average Precision

 Do relevant links appear at the top of the ranked list?  On average what percentage of irrelevant links are seen before

seeing a relevant one? 𝐵𝑄 = σ𝑠=1

𝑂

(𝑄𝑠𝑓𝑑𝑗𝑡𝑗𝑝𝑜 𝑠 × 𝑗𝑡𝑆𝑓𝑚𝑓𝑤𝑏𝑜𝑢 𝑠 ) 𝑆𝑓𝑚𝑓𝑤𝑏𝑜𝑢𝐸𝑝𝑑𝑣𝑛𝑓𝑜𝑢𝑡 𝑄𝑠𝑓𝑑𝑗𝑡𝑗𝑝𝑜 𝑠 = 𝑄𝑠𝑓𝑑𝑗𝑡𝑗𝑝𝑜 𝑝𝑔 𝑚𝑗𝑡𝑢 𝑣𝑞𝑢𝑝 𝑠 𝑗𝑡𝑆𝑓𝑚𝑓𝑤𝑏𝑜𝑢 𝑠 = ቊ1 𝑗𝑔 𝑠 𝑠𝑓𝑚𝑓𝑤𝑏𝑜𝑢 0 𝑝𝑢ℎ𝑓𝑠𝑥𝑗𝑡𝑓 𝑠 = 𝑆𝑏𝑜𝑙 𝑝𝑔 𝑢𝑏𝑠𝑕𝑓𝑢