{"id":26968,"date":"2026-09-27T20:02:12","date_gmt":"2026-09-27T18:02:12","guid":{"rendered":"https:\/\/www.itspy.cz\/thesis\/knowledge-graphs-for-improving-legal-information-retrieval\/"},"modified":"2026-09-27T20:04:52","modified_gmt":"2026-09-27T18:04:52","slug":"knowledge-graphs-for-improving-legal-information-retrieval","status":"publish","type":"thesis","link":"https:\/\/www.itspy.cz\/en\/thesis\/knowledge-graphs-for-improving-legal-information-retrieval\/","title":{"rendered":"Knowledge graphs for improving legal information retrieval"},"content":{"rendered":"<p>Pr\u00e1ca sa zaober\u00e1 n\u00e1vrhom a implement\u00e1ciou syst\u00e9mu na<br \/>\nvyh\u013ead\u00e1vanie pr\u00e1vnych inform\u00e1ci\u00ed zo slovenskej legislat\u00edvy<br \/>\ns vyu\u017eit\u00edm znalostn\u00fdch grafov a jazykov\u00fdch modelov.<br \/>\nMotiv\u00e1ciou pre zvolen\u00fd<br \/>\npr\u00edstup s\u00fa obmedzenia klasick\u00fdch syst\u00e9mov<br \/>\nRetrieval-Augmented Generation, ktor\u00e9 pri pr\u00e1ci s<br \/>\npr\u00e1vnymi textami nedok\u00e1\u017eu spo\u013eahlivo zachyti\u0165 hierarchick\u00e9<br \/>\na s\u00e9mantick\u00e9 vz\u0165ahy medzi pr\u00e1vnymi<br \/>\npojmami a ich zdrojmi. Pr\u00e1ca navrhuje hybridn\u00fd syst\u00e9m Graph<br \/>\nRAG, ktor\u00fd kombinuje grafov\u00fa<br \/>\ndatab\u00e1zu Neo4j obsahuj\u00facu tezaurus slovensk\u00fdch pr\u00e1vnych<br \/>\npojmov s vektorovou datab\u00e1zou FAISS<br \/>\nindexuj\u00facou textov\u00fd korpus pr\u00e1vnych predpisov. Znalostn\u00fd<br \/>\ngraf bol zostaven\u00fd z d\u00e1t siedmich<br \/>\ntematick\u00fdch pr\u00e1vnych tezaurov pokr\u00fdvaj\u00facich oblasti<br \/>\nob\u010dianskeho, obchodn\u00e9ho, trestn\u00e9ho,<br \/>\nfinan\u010dn\u00e9ho, \u00fastavn\u00e9ho, spr\u00e1vneho a pracovn\u00e9ho pr\u00e1va, pri\u010dom<br \/>\nspolu obsahuje 2 781 pr\u00e1vnych<br \/>\npojmov s ich defin\u00edciami, vz\u00e1jomn\u00fdmi vz\u0165ahmi a odkazmi na<br \/>\nkonkr\u00e9tne pr\u00e1vne predpisy. Syst\u00e9m<br \/>\nimplementuje trojvrstvov\u00fa architekt\u00faru pozost\u00e1vaj\u00facu z<br \/>\nvrstvy retrieval, vrstvy reasoning a vrstvy<br \/>\ngener\u00e1cie. Vrstva retrieval vykon\u00e1va hybridn\u00e9 vyh\u013ead\u00e1vanie<br \/>\nnad grafovou datab\u00e1zou pomocou<br \/>\njazyka Cypher a nad vektorovou datab\u00e1zou pomocou<br \/>\nembeddingov\u00e9ho modelu multilingual-e5-<br \/>\nlarge, pri\u010dom rozhodovacia logika na z\u00e1klade hodn\u00f4t istoty<br \/>\nur\u010duje, ktor\u00fd zdroj inform\u00e1ci\u00ed bude pre<br \/>\ndan\u00fd dopyt pou\u017eit\u00fd. Vrstva reasoning zabezpe\u010duje<br \/>\norchestr\u00e1ciu cel\u00e9ho procesu prostredn\u00edctvom<br \/>\nagenta, ktor\u00fd na z\u00e1klade v\u00fdsledkov retrieval vrstvy<br \/>\nformuluje vstup pre jazykov\u00fd model Google<br \/>\nGemini. Syst\u00e9m bol vyhodnoten\u00fd na s\u00fabore 110 pr\u00e1vnych<br \/>\not\u00e1zok pokr\u00fdvaj\u00facich r\u00f4zne oblasti<br \/>\nslovensk\u00e9ho pr\u00e1va, pri\u010dom hodnotenie vykonala pr\u00e1vni\u010dka.<br \/>\nCelkov\u00e1 \u00faspe\u0161nos\u0165 syst\u00e9mu dosiahla<br \/>\n57,3 % \u00faplne spr\u00e1vnych odpoved\u00ed. Pri ot\u00e1zkach, kde syst\u00e9m<br \/>\naktivoval grafov\u00fa cestu vyh\u013ead\u00e1vania,<br \/>\ndosiahol 65,9 % spr\u00e1vnych odpoved\u00ed, zatia\u013e \u010do pri ot\u00e1zkach<br \/>\nspracovan\u00fdch v\u00fdlu\u010dne prostredn\u00edctvom<br \/>\nvektorovej datab\u00e1zy klesla \u00faspe\u0161nos\u0165 na 32,1 %. V\u00fdsledky<br \/>\npotvrdzuj\u00fa, \u017ee integr\u00e1cia znalostn\u00e9ho<br \/>\ngrafu do RAG pipeline merate\u013ene zlep\u0161uje kvalitu odpoved\u00ed<br \/>\nna pr\u00e1vne ot\u00e1zky a zvy\u0161uje spo\u013eahlivos\u0165<br \/>\ncitovania zdrojov.<\/p>\n","protected":false},"featured_media":26700,"template":"","meta":{"_acf_changed":false,"_links_to":"","_links_to_target":""},"university":[467,181],"thesis-year":[466],"class_list":["post-26968","thesis","type-thesis","status-publish","has-post-thumbnail","hentry","thesis-year-466"],"acf":{"autor":"Matej Scislak","portret":"","vedouci":"doc. Ing. Daniel Hl\u00e1dek, PhD."},"_links":{"self":[{"href":"https:\/\/www.itspy.cz\/en\/wp-json\/wp\/v2\/thesis\/26968","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.itspy.cz\/en\/wp-json\/wp\/v2\/thesis"}],"about":[{"href":"https:\/\/www.itspy.cz\/en\/wp-json\/wp\/v2\/types\/thesis"}],"version-history":[{"count":1,"href":"https:\/\/www.itspy.cz\/en\/wp-json\/wp\/v2\/thesis\/26968\/revisions"}],"predecessor-version":[{"id":27023,"href":"https:\/\/www.itspy.cz\/en\/wp-json\/wp\/v2\/thesis\/26968\/revisions\/27023"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.itspy.cz\/en\/wp-json\/wp\/v2\/media\/26700"}],"wp:attachment":[{"href":"https:\/\/www.itspy.cz\/en\/wp-json\/wp\/v2\/media?parent=26968"}],"wp:term":[{"taxonomy":"university","embeddable":true,"href":"https:\/\/www.itspy.cz\/en\/wp-json\/wp\/v2\/university?post=26968"},{"taxonomy":"thesis-year","embeddable":true,"href":"https:\/\/www.itspy.cz\/en\/wp-json\/wp\/v2\/thesis-year?post=26968"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}