Extended Keyword Index Improved Search for Semantic

Extended Keyword Index & Improved Search for Semantic e-Catalog 이동주 IDS 1

Contents u Motivation u Semantic e-Catalog u Search In e-Catalog u Search Strategy u Keyword Index u Scoring Fucntion u Cat. Ont u Conclusion & Future Work IDS 2

Motivation u Keyword l l e-Catalog take a very important role in e-Business many people want to search product information using simple keyword u Semantic l l l Search e-Catalog legacy e-Catalog couldn’t fully express the various and complex product information and relationship semantic e-Catalog system needs suitable search strategy needs IDS 3

Semantic e-Catalog (1) Classification Scheme 1 Classification Scheme 2 …… …… Attribute Product Data Classification Scheme 3 …… v v P 1 … P 4 v v P 2 P 3 v P 4 …… IDS 4

Semantic e-Catalog (2) EC = {E, R}, E = {P, C, A, U} ME ∈ {C, A, U}, MA = {α 1, α 2, . . . , αm} me = {(α, v)| α ∈ MA, v ∈ VALUE} p = { (a, v)| a ∈ A, v ∈ VALUE} R = { (e 1, e 2, r)| e 1 ∈ E 1, e 2 ∈ E 2, E 1 ∈ E, E 2 ∈ E, r ∈ DR} EC : Electronic Catalog E : Entity R : Relationship DR : Definition of Relationship ME : Meta Entity, MA : Meta Attribute P : Product , C : Classification Scheme A : Attribute, U : Unit Of Measure IDS 5

Search In e-Catalog Search Query Search Engine Query Analyzer Sorted List Ranker DB Interface e-Catalog DB IDS 6

Search Strategy u use simple keyword u use semantics implied in e-Catalog l l l relationship between entities construct keyword index of entity’s information (values of attributes) construct extended keyword index with tagging u use l semantics implied in search query extract useful keyword and tag meaning IDS 7

Extended Keyword Index u extended l keyword (voc, tag 1, tag 2, …, tagt) u extend the definition of semantic e-Catalog with extended keyword index e = { (a, v)| a ∈ ATT, v ∈ VALUE} if e is Product ATT is A else ATT is MA ivoc = (voc, tag 1, tag 2, …, tagt) tag 1 is a’s identifier e = {ivoc 1, ivoc 2, …, ivocv} VOC : Vocabulary IDS 8

RDB Structure for Semantic e-Catalog DB Attribute Classification Scheme G 2 B Attribute Group UOM Classification Scheme GUNGB UOM Group Product (Com. Att) Product (Ind. Att) Classification Scheme UNSPSC VOC IDS 9

Extracting Keyword Indexes u different extracting mechanism according to attributes l l name description numeral just use original IDS 10

Process of Keyword Index Extraction use KLT module Analyze Morpheme Structure Select possible result Extends the word using dictionaries it’s different according to attribute Eliminate the useless word Count frequency and mark order Eliminate duplicated word Do tagging and return Keyword List IDS 11

Tags attribute identifier klt_patn word pattern klt_pos types of stem klt_pos 2 normal types of stem klt_josa klt_eomi domain composed indicate how ivoc was composed & extended k_idx order of the ivoc in original v k_cnt total num extended ivoc from original v frequency of voc in original v IDS 12

Scoring Function Score(Q, e) extend the query Q = {q 1, q 2, …, qi, …, qn} qi = {voc, tag 1, tag 2, …, tags} from extended definition with extended keyword index e = {ivoc 1, ivoc 2, …, ivoca} Score(Q, e) = ∑I, j. Score(qi, ivocj) generalize with relationship r related e Score(Q, e) = ∑I, j. Score(qi, ivocj) + ∑k, lwrk*Score(Q, e’l) wrk : weight of relation rk e’l : related entity using rk Score(qi, ivocj), wr dominate total score IDS 13

Cat. Ont u Parser u Loader l easily extensible semi-automated loading tool using XML specification u Searcher l not implemented yet IDS 14

Loading Process Specification - Entity Converting IDS 15

Loading Process Specification - Relationship Converting IDS 16

Loading Process Specification - Keyword Index Construction (1) IDS 17

Loading Process Specification - Keyword Index Construction (2) IDS 18

Conclusion & Future Work u Conclusion l l l u propose extended keyword index using various tag for semantic e. Catalog implement semi-automated converting tool from legacy e-Catalog to semantic e-Catalog with easily extensible XML specification propose scoring function which extended keyword index is applicable Future work l l l contrive feasible scoring function and methods to assign weights of each relationship implement Searcher extend this motel to general E-R model IDS 19