根據(jù)提供的資源數(shù)據(jù)，2017年大數(shù)據(jù)開(kāi)發(fā)職位的招聘人數(shù)大幅增加。具體來(lái)說(shuō)，2016年大數(shù)據(jù)開(kāi)發(fā)的招聘人數(shù)為5,667人，而2017年則增加到41,831人。因此，2017年大數(shù)據(jù)開(kāi)發(fā)職位比2016年增加了36,164人 (41,831 - 5,667 = 36,164)。

os.environ["LLAMA_CLOUD_API_KEY"] = "xxx"

parser = LlamaParse(
        result_type="json",
        use_vendor_multimodal_model=True,
        vendor_multimodal_model_name="gemini-2.0-flash-001",
        language="ch_sim"
    )
md_result = parser.get_json_result(file_path)
doc_id = md_result[0]["job_id"]
pages = md_result[0]["pages"]
# 下載圖片到本地
parser.get_images(md_result, download_path="data_images")

dataset = []
docs = []
base64_map = {}
for page in pages:
    md = page["md"]
    page_number = page["page"]

    # 查找并上傳對(duì)應(yīng)頁(yè)碼的圖片
    local_image_path = find_image_by_page(
        "data_images", doc_id, page_number)
    with open(local_image_path, "rb") as image_file:
        image_base64 = base64.b64encode(image_file.read()).decode('utf-8')

    # 添加到dataset
    dataset.append({
        'content': md,
        'image_base64': image_base64,
        'page_number': page_number
    })
    docs.append(
        Document(
            page_cnotallow=md,
            metadata={
                "page_number": page_number,
            })
    )
    base64_map[page_number] = image_base64
    
vectorstore = Milvus.from_documents(
    documents=docs,
    embedding=embeddings,
    connection_args={"host": "127.0.0.1", "port": "19530"},
    drop_old=True,  # Drop the old Milvus collection if it exists
    collection_name="collection_ppt",
)

dat = vectorstore.similarity_search(query=question, k=5)

image_base64_list = []
chunk = []
for doc in dat:
    page_number = doc.metadata["page_number"]
    print(page_number)
    image_base64 = base64_map.get(page_number)
    if image_base64:
        image_base64_list.append(image_base64)
        chunk.append(doc.page_content)
        
openai_api_key = os.environ["OPENAI_API_KEY"]  # 替換為你的 OpenAI API Key
headers = {
    "Content-Type": "application/json",
    "Authorization": f"Bearer {openai_api_key}"
}

# 構(gòu)建 messages 內(nèi)容
messages_content = [
    {"type": "text", "text": """The following is the Markdown text and image information parsed in the slide. Markdown text has attempted to convert the relevant charts into tables. Give priority to using picture information to answer questions. Use Markdown text information only when the image cannot be understood.
 
Here is the context:
---------
{context}
---------

Question: {question}
""".format(cnotallow="\n".join(chunk), questinotallow=question)}
]

if image_base64_list:
    # 添加所有檢索到的圖片
    for img_base64 in image_base64_list:
        messages_content.append({
            "type": "image_url",
            "image_url": {
                "url": f"data:image/jpeg;base64,{img_base64}"
            }
        })

payload = {
    "model": "gpt-4.1",
    "messages": [
        {
            "role": "user",
            "content": messages_content
        }
    ],
    "temperature": 0.1,
    "max_tokens": 1000# 根據(jù)需要調(diào)整
}

# 發(fā)送請(qǐng)求到 OpenAI API
try:
    response = requests.post(
        "https://api.openai-proxy.com/v1/chat/completions",
        headers=headers,
        jsnotallow=payload
    )
    response.raise_for_status()  # 檢查請(qǐng)求是否成功

    # 解析并打印結(jié)果
    result = response.json()
    print("OpenAI 分析結(jié)果:")
    print(result["choices"][0]["message"]["content"])

except requests.exceptions.RequestException as e:
    print(f"請(qǐng)求 OpenAI API 失敗: {e}")
    if hasattr(e, 'response') and e.response:
        print(f"錯(cuò)誤詳情: {e.response.text}")