Sample Talks for Organizational Hints Krishnaprasad Thirunarayan Department of Computer Science and Engineering Wright State University Dayton, OH-45435

Overall R&D Agenda Develop semi-automatic techniques for information extraction/retrieval to enable man and machine to complement each other in assimilation of semi-structured, heterogeneous documents => Semantic Web Technologies.

A Modular Approach to Document Indexing and Semantic Search

Goal (What?) Background and Motivation (Why?) Implementation Details (How?) Evaluation and Applications (Why?) Conclusions

Goal

Develop a modular approach to improving effectiveness of searching documents for information Reuse and integrate mature software components

Background and Motivation

Improve recall using information implicit in the English language Improve precision and recall using domain- specific information implicit in the document collection Assist manual content extraction by mapping document phrases to controlled vocabulary terms (domain library)  NSF-SBIR Phases I and II with Cohesia Corp.

Enable extensions Spell check input query Organize search results through grouping  Improve precision thro sense-disambiguation Enable experimentation Investigate empirical relationship between significant eigenvalues in the Singular Value Decomposition (SVD) and the number of document clusters using benchmarks.

Implementation Details (How?)

Tools Used Apache’s Lucene APIs A high-performance, Java text search engine library with smart indexing strategies. WordNet and Java WordNet Library NIST and MathWork’s Java Matrix package (JAMA) for LSI Domain-specific controlled vocabulary for Materials and Process Specs

Jazzy, a Java Open Source Spell- Checker MEDLINE dataset 20-Newsgroups dataset Reuters newswire stories datasets

Architecture of Content-based Indexing and Semantic Search Engine Inverted Document Index LSA Term Matrix Document Indexer Configurer Searcher Query Modifier Highlighter WordNet Output User query Domain Library Inverted DLIndex DL Term Locator Document collection

Evaluation and Application (Why?)

Enhanced search illustrating wildcard pattern and synonym expansion

Matching DL Items; DL Term and its location in the document

Example illustrating skippable group

LSI and Clustering Exploring relationship between the number of significant eigenvalues and the number of document clusters  20-Mini-Newsgroup dataset 2000 postings, 20 groups  Reuters Newswire Stories dataset Used 2000 stories at a time, 70 topics

Conclusions

Useful assistance for manual content extraction from materials and process specs, given the controlled vocabulary In future, this framework / infrastructure can be used for experiments with expressive and context-aware search.

Formalizing and Querying Heterogeneous Documents with Tables

Goal (What?) Background and Motivation (Why?) Implementation Details (How?) Evaluation and Applications (Why?) Conclusions

Goal

Define, embed, and use metadata in semi- structured documents containing tables. Content-oriented/domain-specific metadata of human sensible document Makes explicit semantics of complex data Enables augmentation of an interpretation in a modular fashion.

Heterogeneous Document

Background and Motivation

Embedding metadata improves traceability, thereby facilitating Content Extraction Verification Update

Implementation Details (How?)

XML Technology Document-Centric View: XML is used to annotate documents for use by humans in the realm of document processing and content extraction. Data-Centric View: XML is used as text- based format for information exchange / serialization in the context of Web Services.

Basic idea behind our approach Unify the two views by using XML- elements to materialize abstract syntax, and together with XML attributes and XML element definitions, formalize the content.  Key advantage: Minimizes maintenance of additional data structures to relate original document with its formalization.

Two Concrete Implementations Use Web Services language Water which amalgamates XML Technology with programming language concepts Use XML/XSLT infrastructure

Water-based approach Each annotation reflects the semantics of the text fragment it encloses. The annotated data can be interpreted by viewing it as a function/procedure call in Water. The correspondence between formal parameter and actual argument is position-based. The semantics of annotation is defined in Water as a method definition in a class, separately.

Example Table Thickness (mm) Tensile Strength (ksi) Yield Strength (ksi) 0.50 and under – –

Example of Tagged Table Thickness (mm) Tensile Strength (ksi) Yield Strength (ksi) table and under table table table....

Example of Processing Code /> <set rows= table.rows. />/> …

XML/XSLT-based approach Each annotation reflects the semantics of the text fragment it encloses. To make the annotated data XML compliant, dummy attributes such as one, two, three, … etc are introduced. The correspondence between formal attribute and the actual value is name-based. The semantics is defined by interpreting XML-elements and its XML-attributes via XSLT, separately.

Example of Tagged Table <tableSchema one="Thickness(min)" two="Thickness(max)" three="Tensile Strength“ four="Yield Strength"/>...

XSLT Stylesheets can be used to: Query: to perform table look-ups. Transform: to change units of measure such as from standard SI units to FPS units and vice versa. Format: to display the table in HTML form. Extract: to recover the original table. Verify: to check static semantic constraints on table data values.

Evaluation and Application (Why?)

Advantage Only tabular data in each document is annotated. The annotation definition is factored out as background knowledge. Thus, the semantics of each table type is specified just once outside the document and is reused with different documents containing similar tables.

Disadvantage Both avenues require mature tool support for wide spread adoption. For example, develop MS FrontPage like interface where the Master document is the annotated form, and the user explicitly interacts with/edits only a view of the annotated document, for readability reasons, and has support for export as XML to generate well-formed XML document.

Prolog rendition strengthTableRow( 0, 0.50, 165, 155). strengthTableRow(0.50, 1.00, 160, 150). strengthTableRow(1.00, 1.50, 155, 145).... strengthTable(Thickness, TensileStrength, YieldStrength) :- strengthTableRow(L, U, TensileStrength, YieldStrength), L = Thickness. thicknessToTensileStrength(Thickness, TensileStrength) :- strengthTable(Thickness, TensileStrength, _). thicknessToYieldStrength(Thickness, YieldStrength) :- strengthTable(Thickness, _, YieldStrength). ?- thicknessToYieldStrength(0.6,YS).

Conclusion and Future Work

Develop a catalog of predefined tables, specifying them using Semantic Web formalisms (such as RDF, OWL, etc) and mapping the tabular data into a set of pre- defined tables, possibly qualified. Develop techniques for manual mapping of complex tables into simpler ones: To provide semantics to data. To improve traceability. To facilitate automatic manipulation.

Tailor and improve IE and IR techniques developed in the context of text processing to Semantic Web documents such as in XML, RDF, etc benefiting from additional support from ontologies such as in OWL, etc

Our Related Publications

K. Thirunarayan, A. Berkovich, and D. Sokol, An Information Extraction Approach to Reorganizing and Summarizing Specifications, In: Information and Software Technology Journal, Vol. 47, Issue 4, pp , K. Thirunarayan, On Embedding Machine- Processable Semantics into Documents, In: IEEE Transactions on Knowledge and Data Engineering, Vol. 17, No. 7, pp , July 2005.

Holy Grail Ultimately develop principles, techniques and tools, to author and extract human-readable and machine-comprehensible parts of a document hand in hand, and keep them side by side.