Pogovorni agent v slovenskem jeziku za sistem za upravljanje s človeškimi viri

REPŠE, MATIC (author), Bosnić, Zoran (mentor)

V diplomski nalogi smo se lotili implementacije pogovornega robota v slovenščini, ki služi kot vmesnik za hitro iskanje informacij v zaprtem kadrovskem sistemu. Za implementacijo agenta smo uporabili razčlenjevalnik, razvit pri projektu Sporazumevanje v slovenskem jeziku (2008-2013). Razčlenjen vnosni stavek smo obdelali v programskem jeziku Python in mu poiskali vse stavčne člene. Na podlagi najdenih stavčnih členov smo nato poizkusili razumeti, po katerem podatku sprašuje vnosni stavek. V primerih, kjer smo uspešno našli pomen, smo iskani podatek poiskali v kadrovskem sistemu ter uporabniku vrnili odgovor. Preko razvitega vmesnika lahko uporabniki kadrovskega sistema sprašujejo po podatkih, ki so jim v sklopu sistema na voljo z nekaj kliki. Pri evalvaciji smo si pomagali z ročno skladenjsko in oblikoslovno označenim učnim korpusom, ki smo ga ustvarili v sklopu razvoja. Rezultati so pokazali, da smo iskanje pomena zastavili dobro. Z združitvijo našega učnega korpusa in učnega korpusa ssj500k smo uspešnost iskanja pomena dvignili na 84 %.

• Language: Slovenian
• Keywords: obdelava naravnega jezika, pogovorni agent, HRM
• Work type: Bachelor thesis/paper
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2018
• PID: 20.500.12556/RUL-100116
• Publication date in RUL: 08.03.2018

Secondary language

• Language: English
• Title: Slovene chat agent for a human resources management system

Abstract

In this thesis we attempted to implement a slovene chat agent. The agent would serve as an interface to quickly retrieve data from a closed HRM (human resources management) system. To implement the mentioned agent, we used a slovene parser developed in the scope of the Communication in Slovene project. The parsed input sentence was then processed in Python, where we found all of its sentence elements. Knowing its sentence elements, we then tried to understand, which data it was asking for. In cases where we successfully found the meaning, we searched for the wanted data in the HRM system and then answered back to the user if we found it. Through the developed interface, users of the closed HRM system can ask for information with natural slovene language. We tested the developed agent with the new learning corpus we created during the development. Results showed we set up a good meaning searching algorithm. With the merging of our new learning corpus and the learning corpus ssj500k we raised the success rate of our meaning searching algorithm up to 84 %.

• Keywords: NLP, chat agent, HRM