Grafico RAG con Milvus

L'applicazione diffusa di modelli linguistici di grandi dimensioni evidenzia l'importanza di migliorare l'accuratezza e la pertinenza delle loro risposte. La Retrieval-Augmented Generation (RAG) arricchisce i modelli con basi di conoscenza esterne, fornendo maggiori informazioni contestuali e mitigando problemi come l'allucinazione e la conoscenza insufficiente. Tuttavia, affidarsi esclusivamente a semplici paradigmi RAG ha i suoi limiti, soprattutto quando si tratta di relazioni complesse tra entità e domande multi-hop, dove il modello spesso fatica a fornire risposte accurate.

L'introduzione dei grafi di conoscenza (KG) nel sistema RAG offre una nuova soluzione. I KG presentano le entità e le loro relazioni in modo strutturato, fornendo informazioni di recupero più precise e aiutando RAG a gestire meglio le attività di risposta a domande complesse. KG-RAG è ancora in fase iniziale e non esiste un consenso su come recuperare efficacemente entità e relazioni dai KG o su come integrare la ricerca di similarità vettoriale con le strutture a grafo.

In questo quaderno, introduciamo un approccio semplice ma potente per migliorare notevolmente le prestazioni di questo scenario. Si tratta di un semplice paradigma RAG con recupero a più vie e successivo reranking, ma implementa il Graph RAG in modo logico e raggiunge prestazioni all'avanguardia nella gestione di domande multi-hop. Vediamo come viene implementato.

Prerequisiti

Prima di eseguire questo notebook, assicurarsi di avere installato le seguenti dipendenze:

$ pip install --upgrade --quiet pymilvus numpy scipy langchain langchain-core langchain-openai tqdm

Se si utilizza Google Colab, per abilitare le dipendenze appena installate potrebbe essere necessario riavviare il runtime (fare clic sul menu "Runtime" nella parte superiore dello schermo e selezionare "Restart session" dal menu a discesa).

Utilizzeremo i modelli di OpenAI. È necessario preparare la chiave api OPENAI_API_KEY come variabile d'ambiente.

import os

os.environ["OPENAI_API_KEY"] = "sk-***********"

Importare le librerie e le dipendenze necessarie.

import numpy as np

from collections import defaultdict
from scipy.sparse import csr_matrix
from pymilvus import MilvusClient
from langchain_core.messages import AIMessage, HumanMessage
from langchain_core.prompts import ChatPromptTemplate, HumanMessagePromptTemplate
from langchain_core.output_parsers import StrOutputParser, JsonOutputParser
from langchain_openai import ChatOpenAI, OpenAIEmbeddings
from tqdm import tqdm

Inizializzare l'istanza del client Milvus, l'LLM e il modello di incorporamento.

milvus_client = MilvusClient(uri="./milvus.db")

llm = ChatOpenAI(
    model="gpt-4o",
    temperature=0,
)
embedding_model = OpenAIEmbeddings(model="text-embedding-3-small")

Caricamento dei dati offline

Preparazione dei dati

Utilizzeremo come esempio un nano dataset che introduce la relazione tra la famiglia di Bernoulli e quella di Eulero. Il nano dataset contiene 4 passaggi e un insieme di triplette corrispondenti, dove ogni triplette contiene un soggetto, un predicato e un oggetto. In pratica, è possibile utilizzare qualsiasi approccio per estrarre le triplette dal proprio corpus personalizzato.

nano_dataset = [
    {
        "passage": "Jakob Bernoulli (1654–1705): Jakob was one of the earliest members of the Bernoulli family to gain prominence in mathematics. He made significant contributions to calculus, particularly in the development of the theory of probability. He is known for the Bernoulli numbers and the Bernoulli theorem, a precursor to the law of large numbers. He was the older brother of Johann Bernoulli, another influential mathematician, and the two had a complex relationship that involved both collaboration and rivalry.",
        "triplets": [
            ["Jakob Bernoulli", "made significant contributions to", "calculus"],
            [
                "Jakob Bernoulli",
                "made significant contributions to",
                "the theory of probability",
            ],
            ["Jakob Bernoulli", "is known for", "the Bernoulli numbers"],
            ["Jakob Bernoulli", "is known for", "the Bernoulli theorem"],
            ["The Bernoulli theorem", "is a precursor to", "the law of large numbers"],
            ["Jakob Bernoulli", "was the older brother of", "Johann Bernoulli"],
        ],
    },
    {
        "passage": "Johann Bernoulli (1667–1748): Johann, Jakob’s younger brother, was also a major figure in the development of calculus. He worked on infinitesimal calculus and was instrumental in spreading the ideas of Leibniz across Europe. Johann also contributed to the calculus of variations and was known for his work on the brachistochrone problem, which is the curve of fastest descent between two points.",
        "triplets": [
            [
                "Johann Bernoulli",
                "was a major figure of",
                "the development of calculus",
            ],
            ["Johann Bernoulli", "was", "Jakob's younger brother"],
            ["Johann Bernoulli", "worked on", "infinitesimal calculus"],
            ["Johann Bernoulli", "was instrumental in spreading", "Leibniz's ideas"],
            ["Johann Bernoulli", "contributed to", "the calculus of variations"],
            ["Johann Bernoulli", "was known for", "the brachistochrone problem"],
        ],
    },
    {
        "passage": "Daniel Bernoulli (1700–1782): The son of Johann Bernoulli, Daniel made major contributions to fluid dynamics, probability, and statistics. He is most famous for Bernoulli’s principle, which describes the behavior of fluid flow and is fundamental to the understanding of aerodynamics.",
        "triplets": [
            ["Daniel Bernoulli", "was the son of", "Johann Bernoulli"],
            ["Daniel Bernoulli", "made major contributions to", "fluid dynamics"],
            ["Daniel Bernoulli", "made major contributions to", "probability"],
            ["Daniel Bernoulli", "made major contributions to", "statistics"],
            ["Daniel Bernoulli", "is most famous for", "Bernoulli’s principle"],
            [
                "Bernoulli’s principle",
                "is fundamental to",
                "the understanding of aerodynamics",
            ],
        ],
    },
    {
        "passage": "Leonhard Euler (1707–1783) was one of the greatest mathematicians of all time, and his relationship with the Bernoulli family was significant. Euler was born in Basel and was a student of Johann Bernoulli, who recognized his exceptional talent and mentored him in mathematics. Johann Bernoulli’s influence on Euler was profound, and Euler later expanded upon many of the ideas and methods he learned from the Bernoullis.",
        "triplets": [
            [
                "Leonhard Euler",
                "had a significant relationship with",
                "the Bernoulli family",
            ],
            ["leonhard Euler", "was born in", "Basel"],
            ["Leonhard Euler", "was a student of", "Johann Bernoulli"],
            ["Johann Bernoulli's influence", "was profound on", "Euler"],
        ],
    },
]

Le entità e le relazioni sono costruite come segue:

• L'entità è il soggetto o l'oggetto della terzina, quindi li estraiamo direttamente dalle terzine.
• Qui costruiamo il concetto di relazione concatenando direttamente soggetto, predicato e oggetto con uno spazio in mezzo.

Prepariamo anche un dict per mappare l'id dell'entità all'id della relazione e un altro dict per mappare l'id della relazione all'id del passaggio per un uso successivo.

entityid_2_relationids = defaultdict(list)
relationid_2_passageids = defaultdict(list)

entities = []
relations = []
passages = []
for passage_id, dataset_info in enumerate(nano_dataset):
    passage, triplets = dataset_info["passage"], dataset_info["triplets"]
    passages.append(passage)
    for triplet in triplets:
        if triplet[0] not in entities:
            entities.append(triplet[0])
        if triplet[2] not in entities:
            entities.append(triplet[2])
        relation = " ".join(triplet)
        if relation not in relations:
            relations.append(relation)
            entityid_2_relationids[entities.index(triplet[0])].append(
                len(relations) - 1
            )
            entityid_2_relationids[entities.index(triplet[2])].append(
                len(relations) - 1
            )
        relationid_2_passageids[relations.index(relation)].append(passage_id)

Inserimento dei dati

Creare collezioni Milvus per entità, relazione e passaggio. La collezione di entità e quella di relazioni sono utilizzate come collezioni principali per la costruzione del grafo nel nostro metodo, mentre la collezione di passaggi è utilizzata come confronto ingenuo di recupero RAG o come scopo ausiliario.

embedding_dim = len(embedding_model.embed_query("foo"))


def create_milvus_collection(collection_name: str):
    if milvus_client.has_collection(collection_name=collection_name):
        milvus_client.drop_collection(collection_name=collection_name)
    milvus_client.create_collection(
        collection_name=collection_name,
        dimension=embedding_dim,
        consistency_level="Strong",
    )


entity_col_name = "entity_collection"
relation_col_name = "relation_collection"
passage_col_name = "passage_collection"
create_milvus_collection(entity_col_name)
create_milvus_collection(relation_col_name)
create_milvus_collection(passage_col_name)

Inserire i dati con le loro informazioni di metadati nelle raccolte di Milvus, comprese le raccolte di entità, relazioni e passaggi. Le informazioni sui metadati includono l'id del passaggio e l'id dell'entità o della relazione adiacente.

def milvus_insert(
    collection_name: str,
    text_list: list[str],
):
    batch_size = 512
    for row_id in tqdm(range(0, len(text_list), batch_size), desc="Inserting"):
        batch_texts = text_list[row_id : row_id + batch_size]
        batch_embeddings = embedding_model.embed_documents(batch_texts)

        batch_ids = [row_id + j for j in range(len(batch_texts))]
        batch_data = [
            {
                "id": id_,
                "text": text,
                "vector": vector,
            }
            for id_, text, vector in zip(batch_ids, batch_texts, batch_embeddings)
        ]
        milvus_client.insert(
            collection_name=collection_name,
            data=batch_data,
        )


milvus_insert(
    collection_name=relation_col_name,
    text_list=relations,
)

milvus_insert(
    collection_name=entity_col_name,
    text_list=entities,
)

milvus_insert(
    collection_name=passage_col_name,
    text_list=passages,
)

Inserting: 100%|███████████████████████████████████| 1/1 [00:00<00:00,  1.02it/s]
Inserting: 100%|███████████████████████████████████| 1/1 [00:00<00:00,  1.39it/s]
Inserting: 100%|███████████████████████████████████| 1/1 [00:00<00:00,  2.28it/s]

Interrogazione online

Recupero delle somiglianze

Recuperiamo le topK entità e relazioni simili in base alla query di input da Milvus.

Quando si esegue il recupero delle entità, occorre prima estrarre le entità dal testo della query utilizzando un metodo specifico come il NER (Named-entity recognition). Per semplicità, prepariamo qui i risultati del NER. Se si desidera modificare la query come domanda personalizzata, è necessario modificare l'elenco NER corrispondente alla query. In pratica, è possibile utilizzare qualsiasi altro modello o approccio per estrarre le entità dalla query.

query = "What contribution did the son of Euler's teacher make?"

query_ner_list = ["Euler"]
# query_ner_list = ner(query) # In practice, replace it with your custom NER approach

query_ner_embeddings = [
    embedding_model.embed_query(query_ner) for query_ner in query_ner_list
]

top_k = 3

entity_search_res = milvus_client.search(
    collection_name=entity_col_name,
    data=query_ner_embeddings,
    limit=top_k,
    output_fields=["id"],
)

query_embedding = embedding_model.embed_query(query)

relation_search_res = milvus_client.search(
    collection_name=relation_col_name,
    data=[query_embedding],
    limit=top_k,
    output_fields=["id"],
)[0]

Espandere il sottografo

Utilizziamo le entità e le relazioni recuperate per espandere il sottografo e ottenere le relazioni candidate, per poi unirle nei due modi. Ecco un diagramma di flusso del processo di espansione del sottografo:

Costruiamo una matrice di adiacenza e usiamo la moltiplicazione matriciale per calcolare le informazioni di mappatura di adiacenza entro pochi gradi. In questo modo, possiamo ottenere rapidamente informazioni su qualsiasi grado di espansione.

# Construct the adjacency matrix of entities and relations where the value of the adjacency matrix is 1 if an entity is related to a relation, otherwise 0.
entity_relation_adj = np.zeros((len(entities), len(relations)))
for entity_id, entity in enumerate(entities):
    entity_relation_adj[entity_id, entityid_2_relationids[entity_id]] = 1

# Convert the adjacency matrix to a sparse matrix for efficient computation.
entity_relation_adj = csr_matrix(entity_relation_adj)

# Use the entity-relation adjacency matrix to construct 1 degree entity-entity and relation-relation adjacency matrices.
entity_adj_1_degree = entity_relation_adj @ entity_relation_adj.T
relation_adj_1_degree = entity_relation_adj.T @ entity_relation_adj

# Specify the target degree of the subgraph to be expanded.
# 1 or 2 is enough for most cases.
target_degree = 1

# Compute the target degree adjacency matrices using matrix multiplication.
entity_adj_target_degree = entity_adj_1_degree
for _ in range(target_degree - 1):
    entity_adj_target_degree = entity_adj_target_degree * entity_adj_1_degree
relation_adj_target_degree = relation_adj_1_degree
for _ in range(target_degree - 1):
    relation_adj_target_degree = relation_adj_target_degree * relation_adj_1_degree

entity_relation_adj_target_degree = entity_adj_target_degree @ entity_relation_adj

Prendendo il valore dalla matrice di espansione del grado target, possiamo facilmente espandere il grado corrispondente dall'entità e dalle relazioni recuperate per ottenere tutte le relazioni del sottografo.

expanded_relations_from_relation = set()
expanded_relations_from_entity = set()
# You can set the similarity threshold here to guarantee the quality of the retrieved ones.
# entity_sim_filter_thresh = ...
# relation_sim_filter_thresh = ...

filtered_hit_relation_ids = [
    relation_res["entity"]["id"]
    for relation_res in relation_search_res
    # if relation_res['distance'] > relation_sim_filter_thresh
]
for hit_relation_id in filtered_hit_relation_ids:
    expanded_relations_from_relation.update(
        relation_adj_target_degree[hit_relation_id].nonzero()[1].tolist()
    )

filtered_hit_entity_ids = [
    one_entity_res["entity"]["id"]
    for one_entity_search_res in entity_search_res
    for one_entity_res in one_entity_search_res
    # if one_entity_res['distance'] > entity_sim_filter_thresh
]

for filtered_hit_entity_id in filtered_hit_entity_ids:
    expanded_relations_from_entity.update(
        entity_relation_adj_target_degree[filtered_hit_entity_id].nonzero()[1].tolist()
    )

# Merge the expanded relations from the relation and entity retrieval ways.
relation_candidate_ids = list(
    expanded_relations_from_relation | expanded_relations_from_entity
)

relation_candidate_texts = [
    relations[relation_id] for relation_id in relation_candidate_ids
]

Abbiamo ottenuto le relazioni candidate attraverso l'espansione del sottografo, che saranno ricanalizzate da LLM nella fase successiva.

Ricanalizzazione da parte di LLM

In questa fase, utilizziamo il potente meccanismo di autoattenzione di LLM per filtrare e raffinare ulteriormente l'insieme delle relazioni candidate. Utilizziamo un prompt unico, incorporando la domanda e l'insieme di relazioni candidate nel prompt, e istruiamo LLM a selezionare le relazioni potenziali che potrebbero aiutare a rispondere alla domanda. Dato che alcune domande possono essere complesse, adottiamo l'approccio della catena del pensiero, consentendo a LLM di articolare il suo processo di pensiero nella risposta. La risposta di LLM deve essere in formato json, per una comoda analisi.

query_prompt_one_shot_input = """I will provide you with a list of relationship descriptions. Your task is to select 3 relationships that may be useful to answer the given question. Please return a JSON object containing your thought process and a list of the selected relationships in order of their relevance.

Question:
When was the mother of the leader of the Third Crusade born?

Relationship descriptions:
[1] Eleanor was born in 1122.
[2] Eleanor married King Louis VII of France.
[3] Eleanor was the Duchess of Aquitaine.
[4] Eleanor participated in the Second Crusade.
[5] Eleanor had eight children.
[6] Eleanor was married to Henry II of England.
[7] Eleanor was the mother of Richard the Lionheart.
[8] Richard the Lionheart was the King of England.
[9] Henry II was the father of Richard the Lionheart.
[10] Henry II was the King of England.
[11] Richard the Lionheart led the Third Crusade.

"""
query_prompt_one_shot_output = """{"thought_process": "To answer the question about the birth of the mother of the leader of the Third Crusade, I first need to identify who led the Third Crusade and then determine who his mother was. After identifying his mother, I can look for the relationship that mentions her birth.", "useful_relationships": ["[11] Richard the Lionheart led the Third Crusade", "[7] Eleanor was the mother of Richard the Lionheart", "[1] Eleanor was born in 1122"]}"""

query_prompt_template = """Question:
{question}

Relationship descriptions:
{relation_des_str}

"""


def rerank_relations(
    query: str, relation_candidate_texts: list[str], relation_candidate_ids: list[str]
) -> list[int]:
    relation_des_str = "\n".join(
        map(
            lambda item: f"[{item[0]}] {item[1]}",
            zip(relation_candidate_ids, relation_candidate_texts),
        )
    ).strip()
    rerank_prompts = ChatPromptTemplate.from_messages(
        [
            HumanMessage(query_prompt_one_shot_input),
            AIMessage(query_prompt_one_shot_output),
            HumanMessagePromptTemplate.from_template(query_prompt_template),
        ]
    )
    rerank_chain = (
        rerank_prompts
        | llm.bind(response_format={"type": "json_object"})
        | JsonOutputParser()
    )
    rerank_res = rerank_chain.invoke(
        {"question": query, "relation_des_str": relation_des_str}
    )
    rerank_relation_ids = []
    rerank_relation_lines = rerank_res["useful_relationships"]
    id_2_lines = {}
    for line in rerank_relation_lines:
        id_ = int(line[line.find("[") + 1 : line.find("]")])
        id_2_lines[id_] = line.strip()
        rerank_relation_ids.append(id_)
    return rerank_relation_ids


rerank_relation_ids = rerank_relations(
    query,
    relation_candidate_texts=relation_candidate_texts,
    relation_candidate_ids=relation_candidate_ids,
)

Ottenere i risultati finali

Possiamo ottenere i passaggi finali recuperati dalle relazioni riclassificate.

final_top_k = 2

final_passages = []
final_passage_ids = []
for relation_id in rerank_relation_ids:
    for passage_id in relationid_2_passageids[relation_id]:
        if passage_id not in final_passage_ids:
            final_passage_ids.append(passage_id)
            final_passages.append(passages[passage_id])
passages_from_our_method = final_passages[:final_top_k]

Possiamo confrontare i risultati con il metodo RAG ingenuo, che recupera i passaggi topK basati sull'incorporazione della query direttamente dalla raccolta di passaggi.

naive_passage_res = milvus_client.search(
    collection_name=passage_col_name,
    data=[query_embedding],
    limit=final_top_k,
    output_fields=["text"],
)[0]
passages_from_naive_rag = [res["entity"]["text"] for res in naive_passage_res]

print(
    f"Passages retrieved from naive RAG: \n{passages_from_naive_rag}\n\n"
    f"Passages retrieved from our method: \n{passages_from_our_method}\n\n"
)


prompt = ChatPromptTemplate.from_messages(
    [
        (
            "human",
            """Use the following pieces of retrieved context to answer the question. If there is not enough information in the retrieved context to answer the question, just say that you don't know.
Question: {question}
Context: {context}
Answer:""",
        )
    ]
)

rag_chain = prompt | llm | StrOutputParser()

answer_from_naive_rag = rag_chain.invoke(
    {"question": query, "context": "\n".join(passages_from_naive_rag)}
)
answer_from_our_method = rag_chain.invoke(
    {"question": query, "context": "\n".join(passages_from_our_method)}
)

print(
    f"Answer from naive RAG: {answer_from_naive_rag}\n\nAnswer from our method: {answer_from_our_method}"
)

Passages retrieved from naive RAG: 
['Leonhard Euler (1707–1783) was one of the greatest mathematicians of all time, and his relationship with the Bernoulli family was significant. Euler was born in Basel and was a student of Johann Bernoulli, who recognized his exceptional talent and mentored him in mathematics. Johann Bernoulli’s influence on Euler was profound, and Euler later expanded upon many of the ideas and methods he learned from the Bernoullis.', 'Johann Bernoulli (1667–1748): Johann, Jakob’s younger brother, was also a major figure in the development of calculus. He worked on infinitesimal calculus and was instrumental in spreading the ideas of Leibniz across Europe. Johann also contributed to the calculus of variations and was known for his work on the brachistochrone problem, which is the curve of fastest descent between two points.']

Passages retrieved from our method: 
['Leonhard Euler (1707–1783) was one of the greatest mathematicians of all time, and his relationship with the Bernoulli family was significant. Euler was born in Basel and was a student of Johann Bernoulli, who recognized his exceptional talent and mentored him in mathematics. Johann Bernoulli’s influence on Euler was profound, and Euler later expanded upon many of the ideas and methods he learned from the Bernoullis.', 'Daniel Bernoulli (1700–1782): The son of Johann Bernoulli, Daniel made major contributions to fluid dynamics, probability, and statistics. He is most famous for Bernoulli’s principle, which describes the behavior of fluid flow and is fundamental to the understanding of aerodynamics.']


Answer from naive RAG: I don't know. The retrieved context does not provide information about the contributions made by the son of Euler's teacher.

Answer from our method: The son of Euler's teacher, Daniel Bernoulli, made major contributions to fluid dynamics, probability, and statistics. He is most famous for Bernoulli’s principle, which describes the behavior of fluid flow and is fundamental to the understanding of aerodynamics.

Come si può notare, i passaggi recuperati con il metodo RAG ingenuo non hanno trovato un passaggio di verità, il che ha portato a una risposta sbagliata. I passaggi recuperati con il nostro metodo sono corretti e aiutano a ottenere una risposta accurata alla domanda.