Cesar: Um modelo de recomendação conversacional em português brasileiro

This paper introduces the Cesar model, a novel language model for conversational recommendation. In a landscape where personalized recommendations play a crucial role in widely-used platforms, such as social media and streaming services, addressing the challenge of efficiently capturing user-specific desires remains paramount. Cesar tackles this gap by employing advanced natural language processing and transfer learning techniques, adapting to the context of movie recommendations in Brazilian Portuguese. Unlike traditional approaches relying on implicit user features like likes and watch time, Cesar can understand and respond to more specific queries, such as preferences for genre, duration, and specific cast members. Furthermore, the model addresses the "Cold Start" challenge by leveraging prior knowledge to provide meaningful recommendations from the user’s initial interactions. Developed within the accessible Google Colab environment, Cesar’s design facilitates future adaptations for high availability environments. Results indicate that Cesar presents a promising approach to enhance user experience in conversational recommendation systems, offering improved refinement, dynamic interaction, and explainability in generating recommendation lists. This work significantly contributes to expanding the conversational recommendation theme to new languages, advancing the understanding and application of language models in specific contexts, thereby enriching user-platform interactions.

Carregar modelo

from peft import PeftModel, PeftConfig
from transformers import AutoModelForCausalLM
config = PeftConfig.from_pretrained("matheusrdgsf/cesar-ptbr")
model = AutoModelForCausalLM.from_pretrained("TheBloke/zephyr-7B-beta-GPTQ", revision="gptq-8bit-32g-actorder_True", device_map='auto')
model = PeftModel.from_pretrained(model, "matheusrdgsf/cesar-ptbr")

Inferência

from transformers import GenerationConfig
from transformers import AutoTokenizer
tokenizer_model = AutoTokenizer.from_pretrained('TheBloke/zephyr-7B-beta-GPTQ')
tokenizer_template = AutoTokenizer.from_pretrained('HuggingFaceH4/zephyr-7b-alpha')
generation_config = GenerationConfig(
    do_sample=True,
    temperature=0.1,
    top_p=0.25,
    top_k=0,
    max_new_tokens=512,
    repetition_penalty=1.1,
    eos_token_id=tokenizer_model.eos_token_id,
    pad_token_id=tokenizer_model.eos_token_id,
)
def get_inference(
    text,
    model,
    tokenizer_model=tokenizer_model,
    tokenizer_template=tokenizer_template,
    generation_config=generation_config,
):
    st_time = time.time()
    inputs = tokenizer_model(
        tokenizer_template.apply_chat_template(
            [
                {
                    "role": "system",
                    "content": "Você é um chatbot para indicação de filmes. Responda em português e de maneira educada sugestões de filmes para os usuários.",
                },
                {"role": "user", "content": text},
            ],
            tokenize=False,
        ),
        return_tensors="pt",
    ).to("cuda")
    outputs = model.generate(**inputs, generation_config=generation_config)
    print('inference time:', time.time() - st_time)
    return tokenizer_model.decode(outputs[0], skip_special_tokens=True).split('\n')[-1]
get_inference('Poderia indicar filmes de ação de até 2 horas?', model)

O modelo de cesar-ptbr está disponível para utilização na Huggingface.

O dataset de treinamento re_dial_ptbr está disponível para utilização na Huggingface.