|
1 | 1 | 简体中文 | [English](./README.en.md) |
2 | 2 |
|
3 | | -## 这是一个参考GraphRag的dotnet简易实现 |
| 3 | +## 这是一个参考Text2Sql的dotnet简易实现 |
4 | 4 |
|
5 | | -基于微软在论文中提到的实现思路，执行过程GraphRAG主要实现了如下功能： |
6 | | -- Source Documents → Text Chunks：将源文档分割成文本块。 |
7 | | -- Text Chunks → Element Instances：从每个文本块中提取图节点和边的实例。 |
8 | | -- Element Instances → Element Summaries：为每个图元素生成摘要。 |
9 | | -- Element Summaries → Graph Communities：使用社区检测算法将图划分为社区。 |
10 | | -- Graph Communities → Community Summaries：为每个社区生成摘要。 |
11 | | -- Community Summaries → Community Answers → Global Answer：使用社区摘要生成局部答案，然后汇总这些局部答案以生成全局答案。 |
12 | | - |
13 | | -本项目为demo示例，仅用于学习GraphRAG思路。 |
14 | | - |
15 | | -## 您可以直接在项目中引用NuGet包，或者直接使用本项目提供API服务。 |
16 | | - |
17 | | -出于方便，LLM接口目前只兼容了openai的规范，其他大模型可以考虑使用one-api类的集成产品 |
18 | | - |
19 | | -在appsettings.json配置 |
20 | | - |
21 | | -``` |
22 | | - "GraphOpenAI": { |
23 | | - "Key": "sk-xxx", |
24 | | - "EndPoint": "https://api.antsk.cn/", |
25 | | - "ChatModel": "gpt-4o-mini", |
26 | | - "EmbeddingModel": "text-embedding-ada-002" |
27 | | - }, |
28 | | -"TextChunker": { |
29 | | - "LinesToken": 100, |
30 | | - "ParagraphsToken": 1000 |
31 | | -}, |
32 | | -"GraphDBConnection": { |
33 | | - "DbType": "Sqlite", //PostgreSQL |
34 | | - "DBConnection": "Data Source=graph.db", |
35 | | - "VectorConnection": "graphmem.db", //如果用PostgreSQL，可以和DBConnection一致 |
36 | | - "VectorSize": 1536 //DbType=PostgreSQL时需要设置，sqlite可以不设置 |
37 | | -}, |
38 | | -"GraphSearch": { |
39 | | - "SearchMinRelevance": 0.5, //搜索最小相关性 |
40 | | - "SearchLimit": 3, //向量搜索节点限制个数 |
41 | | - "NodeDepth": 3 ,//检索节点深度 |
42 | | - "MaxNodes": 100 //检索最大节点数 |
43 | | -}, |
44 | | -"GraphSys": { |
45 | | - "RetryCounnt": 2 //重试次数，使用国产模型可能会出现json提取失败，增加重试次数可提高可用性 |
46 | | -} |
47 | | -``` |
48 | | -## 启动项目 |
49 | | -``` |
50 | | -dotnet run --project GraphRag.Net.Web.csproj |
51 | | -``` |
52 | | - |
53 | | -## 启动项目后可以通过 |
54 | | -``` |
55 | | -http://localhost:5000/swagger |
56 | | -``` |
57 | | -## 打开swagger查看接口 |
58 | | - |
59 | | - |
60 | | - |
61 | | -### 也可以使用界面 |
62 | | -``` |
63 | | -http://localhost:5000/ |
64 | | -``` |
65 | | - |
66 | | -打开blazor的UI界面，页面提供了文本导入、文件导入，和问答对话，查看知识图谱等功能 |
67 | | - |
68 | | - |
69 | | - |
70 | | - |
71 | | -## Nuget包使用 |
72 | | -``` |
73 | | -dotnet add package GraphRag.Net |
74 | | -``` |
75 | | -## 为了方便进行提示词调整与修改，SK Plugin我们剥离出了项目，您需要把GraphRag.Net.Web项目中的 graphPlugins目录拷贝到你的项目中，并设置： |
76 | | -[graphPlugins](https://github.com/AIDotNet/GraphRag.Net/tree/main/src/GraphRag.Net.Web/graphPlugins) |
77 | | -``` |
78 | | - <ItemGroup> |
79 | | - <None Include="graphPlugins\**"> |
80 | | - <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory> |
81 | | - </None> |
82 | | - </ItemGroup> |
83 | | -``` |
84 | | - |
85 | | -### 默认配置，使用OpenAI标准接口，在配置了OpenAI的appsettings后可以使用下面代码进行注入 |
86 | | -添加包以后，需要进行配置文件的设置以及依赖注入 |
87 | | -``` |
88 | | -//OpenAI配置 |
89 | | -builder.Configuration.GetSection("GraphOpenAI").Get<GraphOpenAIOption>(); |
90 | | -//文档切片配置 |
91 | | -builder.Configuration.GetSection("TextChunker").Get<TextChunkerOption>(); |
92 | | -//配置数据库链接 |
93 | | -builder.Configuration.GetSection("GraphDBConnection").Get<GraphDBConnectionOption>(); |
94 | | -//系统设置 |
95 | | -builder.Configuration.GetSection("GraphSys").Get<GraphSysOption>(); |
96 | | -
|
97 | | -//注入AddGraphRagNet,注意，需要先注入配置文件，然后再注入GraphRagNet |
98 | | -builder.Services.AddGraphRagNet(); |
99 | | -``` |
100 | | - |
101 | | -### 如果你想接入其他模型，可以参考以下代码,这里抽象了Kernel的实现，你可以自定义实现 |
102 | | -``` |
103 | | -var kernelBuild = Kernel.CreateBuilder(); |
104 | | -kernelBuild.Services.AddKeyedSingleton<ITextGenerationService>("mock-text", new MockTextCompletion()); |
105 | | -kernelBuild.Services.AddKeyedSingleton<IChatCompletionService>("mock-chat", new MockChatCompletion()); |
106 | | -kernelBuild.Services.AddSingleton((ITextEmbeddingGenerationService)new MockTextEmbeddingGeneratorService()); |
107 | | -kernelBuild.Services.AddKeyedSingleton("mock-embedding", new MockTextEmbeddingGeneratorService()); |
108 | | -
|
109 | | -builder.Services.AddGraphRagNet(kernelBuild.Build()); |
110 | | -``` |
111 | | - |
112 | | -#### 此处需要注意，由于导入可能分多次导入，没有在导入时自动调用生成社区和全局信息，需要自己根据实际情况调用生成社区和全局信息 |
113 | | - |
114 | | -``` |
115 | | - await _graphService.GraphCommunitiesAsync(index); |
116 | | - await _graphService.GraphGlobalAsync(index); |
117 | | -``` |
118 | | - |
119 | | - |
120 | | -使用时注入 IGraphService 服务,以下为参考示例代码 |
121 | | -``` |
122 | | -namespace GraphRag.Net.Api.Controllers |
123 | | -{ |
124 | | - [Route("api/[controller]/[action]")] |
125 | | - [ApiController] |
126 | | - public class GraphController(IGraphService _graphService) : ControllerBase |
127 | | - { |
128 | | - /// <summary> |
129 | | - /// 获取所有的索引数据 |
130 | | - /// </summary> |
131 | | - /// <returns></returns> |
132 | | - [HttpGet] |
133 | | - public async Task<IActionResult> GetAllIndex() |
134 | | - { |
135 | | - var graphModel = _graphService.GetAllIndex(); |
136 | | - return Ok(graphModel); |
137 | | - } |
138 | | -
|
139 | | -
|
140 | | - /// <summary> |
141 | | - /// 获取所有的图谱数据 |
142 | | - /// </summary> |
143 | | - /// <param name="index"></param> |
144 | | - /// <returns></returns> |
145 | | - [HttpGet] |
146 | | - public async Task<IActionResult> GetAllGraphs(string index) |
147 | | - { |
148 | | - if (string.IsNullOrEmpty(index)) |
149 | | - { |
150 | | - return Ok(new GraphViewModel()); |
151 | | - } |
152 | | - var graphModel = _graphService.GetAllGraphs(index); |
153 | | - return Ok(graphModel); |
154 | | - } |
155 | | -
|
156 | | -
|
157 | | - /// <summary> |
158 | | - /// 插入文本数据 |
159 | | - /// </summary> |
160 | | - /// <param name="model"></param> |
161 | | - /// <returns></returns> |
162 | | - [HttpPost] |
163 | | - public async Task<IActionResult> InsertGraphData(InputModel model) |
164 | | - { |
165 | | - await _graphService.InsertGraphDataAsync(model.Index, model.Input); |
166 | | - return Ok(); |
167 | | - } |
168 | | -
|
169 | | - /// <summary> |
170 | | - /// 搜索递归获取节点相关的所有边和节点进行图谱对话 |
171 | | - /// </summary> |
172 | | - /// <param name="model"></param> |
173 | | - /// <returns></returns> |
174 | | - [HttpPost] |
175 | | - public async Task<IActionResult> SearchGraph(InputModel model) |
176 | | - { |
177 | | - var result = await _graphService.SearchGraphAsync(model.Index, model.Input); |
178 | | - return Ok(result); |
179 | | - } |
180 | | -
|
181 | | - /// <summary> |
182 | | - /// 通过社区算法检索社区节点进行对话 |
183 | | - /// </summary> |
184 | | - /// <param name="model"></param> |
185 | | - /// <returns></returns> |
186 | | - [HttpPost] |
187 | | - public async Task<IActionResult> SearchGraphCommunity(InputModel model) |
188 | | - { |
189 | | - var result = await _graphService.SearchGraphCommunityAsync(model.Index, model.Input); |
190 | | - return Ok(result); |
191 | | - } |
192 | | -
|
193 | | - /// <summary> |
194 | | - /// 导入txt文档 |
195 | | - /// </summary> |
196 | | - /// <param name="index"></param> |
197 | | - /// <param name="file"></param> |
198 | | - /// <returns></returns> |
199 | | - [HttpPost] |
200 | | - public async Task<IActionResult> ImportTxt(string index,IFormFile file) |
201 | | - { |
202 | | - var forms = await Request.ReadFormAsync(); |
203 | | - using (var stream = new StreamReader(file.OpenReadStream())) |
204 | | - { |
205 | | - var txt = await stream.ReadToEndAsync(); |
206 | | - await _graphService.InsertTextChunkAsync(index,txt); |
207 | | - return Ok(); |
208 | | - } |
209 | | - } |
210 | | -
|
211 | | - /// <summary> |
212 | | - /// 通过社区检测生成社区和摘要 |
213 | | - /// </summary> |
214 | | - /// <param name="index"></param> |
215 | | - /// <returns></returns> |
216 | | - [HttpGet] |
217 | | - public async Task<IActionResult> GraphCommunities(string index) |
218 | | - { |
219 | | - await _graphService.GraphCommunitiesAsync(index); |
220 | | - return Ok(); |
221 | | - } |
222 | | - |
223 | | - /// <summary> |
224 | | - /// 通过社区摘要生成全局摘要 |
225 | | - /// </summary> |
226 | | - /// <param name="index"></param> |
227 | | - /// <returns></returns> |
228 | | - [HttpGet] |
229 | | - public async Task<IActionResult> GraphGlobal(string index) |
230 | | - { |
231 | | - await _graphService.GraphGlobalAsync(index); |
232 | | - return Ok(); |
233 | | - } |
234 | | -
|
235 | | - /// <summary> |
236 | | - /// 删除图谱数据 |
237 | | - /// </summary> |
238 | | - /// <param name="index"></param> |
239 | | - /// <returns></returns> |
240 | | - [HttpGet] |
241 | | - public async Task<IActionResult> DeleteGraph(string index) |
242 | | - { |
243 | | - await _graphService.DeleteGraph(index); |
244 | | - return Ok(); |
245 | | - } |
246 | | - } |
247 | | -
|
248 | | - public class InputModel |
249 | | - { |
250 | | - public string Index { get; set; } |
251 | | - public string Input { get; set; } |
252 | | - } |
253 | | -} |
254 | | -
|
255 | | -``` |
256 | | - |
257 | | -## 测试DB，有社区朋友提前预训练了一些数据，链接如下，下载后直接放进项目目录替换即可测试体验 |
258 | | -``` |
259 | | -https://pan.quark.cn/s/bf2d21f29f85 |
260 | | -``` |
261 | | - |
262 | | -## 更多Rag场景可查看 AntSK |
263 | | -项目地址：[AntSK](https://github.com/AIDotNet/AntSK) |
264 | | - |
265 | | -体验环境： |
266 | | - |
267 | | -[Demo地址](https://demo.antsk.cn) |
268 | | - |
269 | | -账号：test |
270 | | - |
271 | | -密码：test |
272 | 5 |
|
273 | 6 |
|
274 | 7 | 也欢迎大家加入我们的微信交流群，可以添加我的微信：**xuzeyu91** 发送进群 |
0 commit comments