Skip to content

Latest commit

 

History

History
57 lines (51 loc) · 2.67 KB

README.md

File metadata and controls

57 lines (51 loc) · 2.67 KB
Raw

Relationship-Scenario based Trailer Recommendation(RSTR)

PPT
RSTR Demo
RSTR DataBase
RSTR API Test
Paper
Committee member comment

Overview Flow

  • Pretrained Model(I used 3 kind of embeddings method)
  1. BERT(trained: wiki(WordPiece), fine tunning: single article or single storyline)
  2. Word2Vec-SG(Train Data: dcard mood article(Entity), yahoo and pixnet storyline(Entity))
  3. Word2Vec-SG(Train Data: dcard mood article(Word), yahoo and pixnet storyline(Word))
  • Feature Generation
  1. E2V-BERT 透過產生的斷詞進行辭典過濾並得到各個 entity 在進行 relationship feature 和 scenaio feature 的產生
  2. E2V-W2V-SG 透過產生的斷詞進行辭典過濾並得到各個 entity 在進行 relationship feature 和 scenario feature 的產生
  3. W2V-W2V-SG(relationship model baseline) 單純的斷詞未經過辭典產生 word 並做詞向量相加
  • Relationship Classifer
    relationship classifier 訓練(relationship_algorithm_analysis) > 供給 server 存取(main)
  • Scenario Classifier
    scenario classifier 訓練(scenario_algorithm_analysis) > 供給 server 存取(main)

Main

RecommenderSystem

目的：NER 運用(Insert Article, Movie Parser and Article, Movie NER), Server 架設
執行：activator "start 8309"(註：java project)

main

目的：存取模型結果, Server 架設
執行：python3 server.py(註：python project)

token

執行：python3 service_Server.py(註：python project)

Module

main_embedding

目的：訓練 Entity2Vec-BERT, Entity2Vec-W2V-SG, Word2Vec-W2V-SG(baseline) 並將 relationship feature and scenario feature 存入

relationship_algorithm_analysis

目的：訓練 relationship 模型(CNN)

scenario_algorithm_analysis

目的：訓練 scenario 模型(KNN, NB, SVM, RFC)

rstr_evaluation

目的：給予評分項目，產生評估結果

knowledge base

目的：產生 relationship lexicon(person), emotion lexicon(emotion), time lexicon(time), location lexicon(location)and event 辭典

article(Dcard Mood)

目的：爬蟲, 存取資料庫, CKIP Parser

movie(Pixnet and Yahoo)

目的：爬蟲, 存取資料庫, CKIP Parser

test

目的：測試資料用

View

rstr_view

目的：推薦系統介面

rstr_view_evaluation

目的：評估系統介面