210 likes | 310 Views
Logic-based Semantically Enriched Integration of Multi-Feature MIR. Dominik Lübbers Computer Science Department V (Information Systems) Prof. Matthias Jarke RWTH Aachen, Germany. Some Personal Information. Name: Dominik Lübbers
E N D
Logic-based Semantically Enriched Integration of Multi-Feature MIR Dominik Lübbers Computer Science Department V (Information Systems) Prof. Matthias Jarke RWTH Aachen, Germany
Some Personal Information • Name: Dominik Lübbers • Ph.D. student at CS Department V(Information Systems, Prof. Jarke), RWTH Aachen, Germany • main CS research interests: • Multimedia Databases • Information Retrieval • Formal Logics, esp. Description Logics and its applications • Data Mining • Data Quality • Musical interests • Playing piano & organ • part-time church musician • Singing in (mainly chamber) choirs
Motivation • One reason for difficulty of MIR: • User cannot express his information demand easily • Multiple features as „hints“ • Standard metadata (author, title, …) • Melody Similarity (e.g. Query by Humming) • Sound Similarity (e.g. Query By Example) • Lyric fragments („classical“ information retrieval) • … • Queries cannot be understood without respecting (user‘s) background knowledge • Central goal: Formalism and Retrieval Mechanisms to • Integrate similarity-based queries with • Ontology-based Information Retrieval => Intelligent Music Information Retrieval Systems
Overall Approach • Based on previous work done by Umberto Straccia • Relevance description logics • Fuzzy description logics • Ph.D thesis „Foundations of a logic based approach to Multimedia Document Retrieval“ (1999) • Information retrieval as logical inference (van Rijsbergen): • Relevance of for
Overall Approach Form Semantics Piece Instrument playedBy hasRecording class hierarchy Recording Organ Piano Orchestra Int_As Track 1 bwv565 organKreuzkircheBonn Track 2 hasRecording … playedBy MIDI file recSimonPreston hasRecording Track 10 Measure 10-15 recHistoricTelecasts chicagoSymphonyOrchestra playedBy media dependant information OO model media independant information Non-standard DL model
Agenda • Object oriented modelling of media dependant information (form part) • Stepwise development of suitable logic • Standard Description Logics • Inconsistencies and Relevance Logics • Imprecision and Fuzzy Logics • Reasoning about form in • Reasoning about form and semantics • (Some of the many) open questions
Classic Description Logics I • Main purpose: Ontology formalization • Represent relationships between terms in a domain of interest • Many application areas: • Conceptual Data Modelling • Semantic Query Optimization • Software Engineering • Configuration Management • Representation of the meaning of Web resources: Semantic Web (OWL-DL is a Description Logic), … • Good compromise between expressivity and computational complexity • Thoroughly investigated family of logics with many variantsWe concentrate on basic DL
Classic Description Logics II : TBox (terminological knowledge) s : ABox (Assertions, statements about objects in the KB)
Relevance and Inconsistency • Material implication allows for valid sentences, although is not related to • (Inconsistent knowledge base) • Since this piece is by Hensel and by Mendelssohn, Beethoven wrote 11 symphonies • (tautologies independent of premise) • Since Beethoven wrote 11 symphonies,this piece is by Mendelssohn or not. • Relevance Logic: Reject logical inferences where the premise is not relevant to the conclusion
Relevance and Inconsistency II • Avoid „fallacies of relevance“ by four-valued logic • Denotational semantics: • Explicitly and independently interpret falsehood Be aware:
Imprecision and Fuzzy Logic • Semantic model of application domain is imperfect and can contain vague concepts (think of genre…) • Approach: • Replace crisp interpretation of conceptsby fuzzy interpretations: • Fuzzy assertions: • Terminological constraints: • Allows for integration of form-based similarity measures as fuzzy predicates
Horn rules allow basic reasoning with fuzzy n-ary predicates • Specify combination of evidence by nondecreasing function of membership degrees • So far: no recursive rules possible • Queries:
Reasoning about form • OO model of media-dependant data as concrete domain in with fixed interpretation • Represent similarity as fuzzy binary predicate, e.g. • Combine similarity queries, e.g.
Reasoning about form + semantics • Link form and semantics by (fuzzy) predicate Int_As
Some example queries Find all MIDI files that have melody A, contain lyrics B and are transcribed organ pieces Find all MIDI files that additionally sound similar to C (somewhere)
(Some) Open Questions • What parts of the model language do we need for music information? What must be extended? • What are suitable models for the form perspective of music data? What are meaningful ontologies? • How to integrate similarity measures? What interfaces are meaningful? • How to combine membership degrees? • Complexity issues, „query plan evaluation“ • User interfaces for query formulation, …
Modelling the form dimension • Modelling of media-dependent information according to OO-principles • Classify documents by assigning a class with defined attributes (+ types) • Basic class: MDO (media data object) represents linear stream of bytes • Organize classes in a specialization hierarchy • Describe parts of MDOs as CSMO (complex single media objects) by Region-functions • Aggregate CSMO to more complex CSMO • Model similarity measurement as methods in classes
Modelling the form dimension - Example MDO MidiFile CSMO MidiTrack MidiTrack provides method
Classic Description Logics • Syntax • Primitive concepts: (~ unary predicates) • Roles: (~ binary predicates) • Concept term operators • Semantics • Interpretation
Projects and Competences at i5 ConceptBase SEWASIE ProLearn CRC 427: Media&Cultural Communication SFB 476 Improve PRIME DWQ MEMO • Deductive Databases • Conceptual Modelling • Formal Logics, esp. Description Logics • Multimedia Data Management, esp. MPEG-7 • Service-oriented Information Systems • Ontology Engineering • Semantic Web • Data Quality • Data Mining • E-Commerce, esp. Electronic Marketplaces