beyond_vector_search

超越矢量搜索
用於增強數據檢索的混合矢量/知識圖查詢引擎

安裝

我們使用Python == 3.9.18，建議使用虛擬環境安裝所需的軟件包。

 pip install -r requirements.txt

創建過濾數據

• notebooks/parsing_json.ipynb的過濾數據
• notebooks/parsing_cnn_news.ipynb新聞的過濾數據
• notebooks/parsing_wiki_movies.ipynb電影的過濾數據

數據集

我們從Kaggle使用了ARXIV數據集的一個子集，其中包含12,926個樣本，並且醃製文件提供在“ hewer_vector_search/data/fielded_data.pickle”上。

原始碼

 make_vectordb.py: a script to build a vector database from a "data/filtered_data.pickle"

utils/
    - build_graph.py: a script containing helper functions for building the knowledge graph
    - parse_arxiv.py: a script containing helper functions for parsing the arxiv dataset
vector_graph/
    - bipartite_graph_dict.py: A custom implementation of the bipartite graph
    - bipartite_graph_networkx.py: An experimental implementation of the bipartite graph using networkx
    - embedding_models.py: A custom implementation of the embedding models for generating the text embeddings
workloads
    - keyword_extractor.py
    - query_gen.py: A script for generating the text queries given paper data points
    - workload_gen.sh: This is the script for generating the workloads we described in the report
testing
    - inference.py: A script for executing our various search query engines on the generated workloads
zy_testing
    - compute_metrics_cos.py: A script for computing the accuracy of our results utilizing various performance compute_metrics_cos