From cfabd60001d41d7193a1e81f2e380061b682d22c Mon Sep 17 00:00:00 2001
From: huangyulin <hyl97@outlook.com>
Date: Fri, 28 Apr 2023 22:18:53 +0800
Subject: [PATCH] =?UTF-8?q?=E6=96=B0=E5=A2=9E=EF=BC=9A=E6=96=87=E6=9C=AC?=
 =?UTF-8?q?=E6=A6=82=E6=8B=AC?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 content/文本概括 Summarizing.ipynb | 82 +++++++++++++++---------------
 1 file changed, 41 insertions(+), 41 deletions(-)

diff --git a/content/文本概括 Summarizing.ipynb b/content/文本概括 Summarizing.ipynb
index 9e5b7db..b40c397 100644
--- a/content/文本概括 Summarizing.ipynb	
+++ b/content/文本概括 Summarizing.ipynb	
@@ -2,7 +2,7 @@
  "cells": [
   {
    "cell_type": "markdown",
-   "id": "716d29bd",
+   "id": "b58204ea",
    "metadata": {},
    "source": [
     "# 文本概括 Summarizing"
@@ -10,7 +10,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "f6b7fca0",
+   "id": "b70ad003",
    "metadata": {},
    "source": [
     "## 1 引言"
@@ -18,7 +18,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "4b819ee9",
+   "id": "12fa9ea4",
    "metadata": {},
    "source": [
     "当今世界上有太多的文本信息，几乎没有人能够拥有足够的时间去阅读所有我们想了解的东西。但令人感到欣喜的是，目前LLM在文本概括任务上展现了强大的水准，也已经有不少团队将这项功能插入了自己的软件应用中。\n",
@@ -28,7 +28,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "48b35259",
+   "id": "1de4fd1e",
    "metadata": {},
    "source": [
     "首先，我们需要OpenAI包，加载API密钥，定义getCompletion函数。"
@@ -37,7 +37,7 @@
   {
    "cell_type": "code",
    "execution_count": 1,
-   "id": "5ab873ff",
+   "id": "9f679f1f",
    "metadata": {},
    "outputs": [],
    "source": [
@@ -56,7 +56,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "469f1276",
+   "id": "9cca835b",
    "metadata": {},
    "source": [
     "## 2 单一文本概括Prompt实验"
@@ -64,7 +64,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "cd7d8580",
+   "id": "0c1e1b92",
    "metadata": {},
    "source": [
     "这里我们举了个商品评论的例子。对于电商平台来说，网站上往往存在着海量的商品评论，这些评论反映了所有客户的想法。如果我们拥有一个工具去概括这些海量、冗长的评论，便能够快速地浏览更多评论，洞悉客户的偏好，从而指导平台与商家提供更优质的服务。"
@@ -72,7 +72,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "2fca0e2b",
+   "id": "9dc2e2bc",
    "metadata": {},
    "source": [
     "**输入文本**"
@@ -81,7 +81,7 @@
   {
    "cell_type": "code",
    "execution_count": 2,
-   "id": "3d3bd87d",
+   "id": "4d9c0eeb",
    "metadata": {},
    "outputs": [],
    "source": [
@@ -99,7 +99,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "1ec202f1",
+   "id": "aad5bd2a",
    "metadata": {},
    "source": [
     "**输入文本（中文翻译）**"
@@ -108,7 +108,7 @@
   {
    "cell_type": "code",
    "execution_count": 3,
-   "id": "84a653de",
+   "id": "43b5dd25",
    "metadata": {},
    "outputs": [],
    "source": [
@@ -122,7 +122,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "e0f78610",
+   "id": "662c9cd2",
    "metadata": {},
    "source": [
     "### 2.1 限制输出文本长度"
@@ -130,7 +130,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "4066804b",
+   "id": "a6d10814",
    "metadata": {},
    "source": [
     "我们尝试限制文本长度为最多30词。"
@@ -139,7 +139,7 @@
   {
    "cell_type": "code",
    "execution_count": 4,
-   "id": "d6e423f2",
+   "id": "02208fbc",
    "metadata": {},
    "outputs": [
     {
@@ -167,7 +167,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "3586d82d",
+   "id": "0df0eb90",
    "metadata": {},
    "source": [
     "中文翻译版本"
@@ -176,7 +176,7 @@
   {
    "cell_type": "code",
    "execution_count": 5,
-   "id": "51cbed99",
+   "id": "bf4b39f9",
    "metadata": {},
    "outputs": [
     {
@@ -202,7 +202,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "08c1643c",
+   "id": "e9ab145e",
    "metadata": {},
    "source": [
     "### 2.2 关键角度侧重"
@@ -210,7 +210,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "f2582b5f",
+   "id": "f84d0123",
    "metadata": {},
    "source": [
     "有时，针对不同的业务，我们对文本的侧重会有所不同。例如对于商品评论文本，物流会更关心运输时效，商家更加关心价格与商品质量，平台更关心整体服务体验。\n",
@@ -220,7 +220,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "9da7a497",
+   "id": "d6f8509a",
    "metadata": {},
    "source": [
     "**侧重于运输**"
@@ -229,7 +229,7 @@
   {
    "cell_type": "code",
    "execution_count": 6,
-   "id": "1432a7fe",
+   "id": "9d8a32a6",
    "metadata": {},
    "outputs": [
     {
@@ -259,7 +259,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "d3b46d0f",
+   "id": "0bd4243a",
    "metadata": {},
    "source": [
     "中文翻译版本"
@@ -268,7 +268,7 @@
   {
    "cell_type": "code",
    "execution_count": 8,
-   "id": "8fff34de",
+   "id": "80636c3e",
    "metadata": {},
    "outputs": [
     {
@@ -294,7 +294,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "426fb25f",
+   "id": "76c97fea",
    "metadata": {},
    "source": [
     "可以看到，输出结果以“快递提前一天到货”开头，体现了对于快递效率的侧重。"
@@ -302,7 +302,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "aaa71480",
+   "id": "83275907",
    "metadata": {},
    "source": [
     "**侧重于价格与质量**"
@@ -311,7 +311,7 @@
   {
    "cell_type": "code",
    "execution_count": 9,
-   "id": "18546578",
+   "id": "767f252c",
    "metadata": {},
    "outputs": [
     {
@@ -342,7 +342,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "63645a57",
+   "id": "cf54fac4",
    "metadata": {},
    "source": [
     "中文翻译版本"
@@ -351,7 +351,7 @@
   {
    "cell_type": "code",
    "execution_count": 12,
-   "id": "826ecc6d",
+   "id": "728d6c57",
    "metadata": {},
    "outputs": [
     {
@@ -377,7 +377,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "d181a9b4",
+   "id": "972dbb1b",
    "metadata": {},
    "source": [
     "可以看到，输出结果以“质量好、价格小贵、尺寸小”开头，体现了对于产品价格与质量的侧重。"
@@ -385,7 +385,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "57595596",
+   "id": "b3ed53d2",
    "metadata": {},
    "source": [
     "### 2.3 关键信息提取"
@@ -393,7 +393,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "5c1fe1d5",
+   "id": "ba6f5c25",
    "metadata": {},
    "source": [
     "在2.2节中，虽然我们通过添加关键角度侧重的Prompt，使得文本摘要更侧重于某一特定方面，但是可以发现，结果中也会保留一些其他信息，如价格与质量角度的概括中仍保留了“快递提前到货”的信息。有时这些信息是有帮助的，但如果我们只想要提取某一角度的信息，并过滤掉其他所有信息，则可以要求LLM进行“文本提取(Extract)”而非“文本概括(Summarize)”。"
@@ -402,7 +402,7 @@
   {
    "cell_type": "code",
    "execution_count": 13,
-   "id": "8ae5c70f",
+   "id": "2d60dc58",
    "metadata": {},
    "outputs": [
     {
@@ -432,7 +432,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "149b594e",
+   "id": "0339b877",
    "metadata": {},
    "source": [
     "中文翻译版本"
@@ -440,21 +440,21 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 16,
-   "id": "5993dd50",
+   "execution_count": 19,
+   "id": "c845ccab",
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "快递提前一天到货，但产品有点小，相比价钱来说可以买到更大的。\n"
+      "快递比预期提前了一天到货。\n"
      ]
     }
    ],
    "source": [
     "prompt = f\"\"\"\n",
-    "你的任务是从电子商务网站上的产品评论中提取相关信息，以向运输部门提供反馈。\n",
+    "你的任务是从电子商务网站上的产品评论中提取相关信息。\n",
     "\n",
     "请从以下三个反引号之间的评论文本中提取产品运输相关的信息，最多30个词汇。\n",
     "\n",
@@ -467,7 +467,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "9fd0c3b8",
+   "id": "50498a2b",
    "metadata": {},
    "source": [
     "## 3 多条文本概括Prompt实验"
@@ -475,7 +475,7 @@
   },
   {
    "cell_type": "markdown",
-   "id": "2b27575c",
+   "id": "a291541a",
    "metadata": {},
    "source": [
     "在实际的工作流中，我们往往有许许多多的评论文本，以下展示了一个基于for循环调用“文本概括”工具并依次打印的示例。当然，在实际生产中，对于上百万甚至上千万的评论文本，使用for循环也是不现实的，可能需要考虑整合评论、分布式等方法提升运算效率。"
@@ -484,7 +484,7 @@
   {
    "cell_type": "code",
    "execution_count": 17,
-   "id": "74f9930d",
+   "id": "ee7caa78",
    "metadata": {},
    "outputs": [],
    "source": [
@@ -564,7 +564,7 @@
   {
    "cell_type": "code",
    "execution_count": null,
-   "id": "5fb23db5",
+   "id": "9d1aa5ac",
    "metadata": {},
    "outputs": [],
    "source": [
@@ -586,7 +586,7 @@
   {
    "cell_type": "code",
    "execution_count": null,
-   "id": "c0be2824",
+   "id": "eb878522",
    "metadata": {},
    "outputs": [],
    "source": []