Azure Durable Function for Text Analysis

A Simple Azure Durable Function Wrapper for Azure Text Analysis

For Azure Cognitive Text Analysis services, there is a limitation for 5,120 characters. With this restriction, Text Analysis Services are hardly useful for anything longer than a couple of pages. This wrapper, based on Azure Durable Function, gets rid of this obstacle by converting Text Analysis into async services.
For the 2022-10-01-preview version, Cognitive Service supports async operations for a limit of 125,000 characters; most of the available async operations have been implemented now. Therefore, we can have far fewer transactions, therefore it will be way lower in cost!
Another common challeng is how to deal with multiple languages in Translator services, there is a rather straight-forward approach, not involving chunking into lines or sentences. It involves two steps.

1. Use language detect to include all alternatives for minor languages.
2. Loop through all languages, regardless of major or minor languages, and specify "from" language parameter. I will spend some time to have this implemented shortly.

Description of Technical Approach

At the core of the Azure Durable Function,

1. a piece of long text is chopped into multiple small chunks by parapragh(s) and delimitors you selected
2. chunks are sent to Azure Text Analysis Services for processing piece by piece (the code is tested on a very limited Azure resource, so all calls are sequential, if your resources allow, you can easily rewrite this into a fan-out and fan-in parallel model)
3. service results from each chunk are combined / merged into a final result

Setup

In order to set this up

1. download the source codes
2. complie into binary with VS2022
3. publish into Azure as a durable funcion with .Net6 stack (no other configurations are needed)

Usage

To create a test console application

1. init a http client
2. add headers
   • Ocp-Apim-Subscription-Key (required)
   • Ocp-Apim-Subscription-Region (required)
   • Ocp-Apim-Subscription-Url (required)
   • Ocp-Apim-Subscription-Method (required)
   • Ocp-Apim-Subscription-Chunk-Size (optional - default: 5,000 for sync operation 120,000 for async operation)
   • Ocp-Apim-Subscription-Splitors (optional - character return and line feed are always there)
   • Ocp-Apim-Subscription-Is-Async (optional - default: false)
3. hash out a json body with a long string variable
4. point to Azure Durable Function entry point
5. constantly query Azure Status Uri
6. when complete, read results

Sample Test Codes

If you want to detect multiple langauges, you might want to specify a smaller chunk size; due to the fact that one chunk can only return one major language, e.g. English is 51% and Japanense is 49%, the service will only return one value for English.
If your test has some chunking issues with paragraph only, e.g. 5,000 characters without \r\n, you might want to specify extra delimitors. Based on HTTP Header protocol, if multiple delimitors are defined, please separate them by ",", e.g. "!,?,.".
When translation service or is translation service for language detection is invoked, the chunk size will set up to 50K characters by system, your custom delimitors are still applied.

using System.Net.Http.Headers;
using System.Text;
using System;
using System.Text.Json.Serialization;
using Newtonsoft.Json;
using Newtonsoft.Json.Linq;

namespace Tester
{
    internal class Program
    {
        static HttpClient _httpClient = new HttpClient();
        static void Main(string[] args)
        {
            string longtxt = File.ReadAllText("somelongtextfile.txt");
            
            
            HttpRequestMessage msg = new HttpRequestMessage();
            msg.Method = HttpMethod.Post;
            msg.RequestUri = new Uri("https://yourazuredureablefunction.azurewebsites.net/api/TextAnalysis_HttpStart?code=yourfunctionaccesscode");
            msg.Headers.Add("Ocp-Apim-Subscription-Key", "yourkeyfromcognitiveservices");
            msg.Headers.Add("Ocp-Apim-Subscription-Region", "yourresourceregion");
            msg.Headers.Add("Ocp-Apim-Subscription-Url", "https://eastus2.api.cognitive.microsoft.com/language/analyze-text/jobs?api-version=2022-10-01-preview");
            msg.Headers.Add("Ocp-Apim-Subscription-Method", "ExtractiveSummarization");
            msg.Headers.Add("Ocp-Apim-Subscription-Chunk-Size", "3000");

            object data = new
            {
                displayName = "Document Summarization Task Example",
                analysisInput = new
                {
                    documents = new[] { new { id = 1, language = "en", text = longtxt } }
                },
                tasks = new[]
                {
                    new
                    {
                        kind = "ExtractiveSummarization",
                        taskName = "Document Summarization Task",
                        parameters = new { sentenceCount = 3 }
                    }
                }
            };

            StringContent content = new StringContent(JsonConvert.SerializeObject(data), Encoding.UTF8, "application/json");
            content.Headers.ContentType = new MediaTypeHeaderValue("application/json");
            msg.Content = content;

            HttpResponseMessage response = _httpClient.Send(msg);
            string resp = response.Content.ReadAsStringAsync().GetAwaiter().GetResult();
            dynamic durableDyn = JsonConvert.DeserializeObject<dynamic>(resp);

            string statusUrl = durableDyn["statusQueryGetUri"];

            HttpRequestMessage statusMsg = new HttpRequestMessage();
            statusMsg.Method = HttpMethod.Get;
            statusMsg.RequestUri = new Uri(statusUrl);

            HttpResponseMessage statusResponse = _httpClient.Send(statusMsg);
            string statusResp = statusResponse.Content.ReadAsStringAsync().GetAwaiter().GetResult();
            dynamic statusDyn = JsonConvert.DeserializeObject<dynamic>(statusResp);

            while (statusDyn["runtimeStatus"] != "Completed")
            { 
                Thread.Sleep(TimeSpan.FromSeconds(10));

                statusMsg = new HttpRequestMessage();
                statusMsg.Method = HttpMethod.Get;
                statusMsg.RequestUri = new Uri(statusUrl);

                statusResponse = _httpClient.Send(statusMsg);
                statusResp = statusResponse.Content.ReadAsStringAsync().GetAwaiter().GetResult();
                statusDyn = JsonConvert.DeserializeObject<dynamic>(statusResp);
            }

            JArray outputs = statusDyn["output"];
            foreach (JValue output in outputs)
            { 
                Console.WriteLine(output.ToString());
            }
        }
    }
}

Another sample for summarization service with async operation.
The following test codes are getting extractive summariztion from a long text file.
Based my test, aroung 15K characters take about one minute to fihish. Compared with other services, Summarization is particular show, for example, Entity Linking can finish 750K charcaters in around 40 seconds. As of writing, the Substrative Summarization is still a gated preview, so you will need submit a request form for access.
The Extractive Summarization seems only work in 2022-10-01-preview version with async operation.\

using Newtonsoft.Json;
using Newtonsoft.Json.Linq;
using System.Diagnostics;
using System.Net.Http.Headers;
using System.Text;

namespace Tester
{
    internal class Program
    {
        static HttpClient _httpClient = new HttpClient();
        static void Main(string[] args)
        {
            string longtxt = File.ReadAllText("news_text.txt");

            HttpRequestMessage msg = new HttpRequestMessage();
            msg.Method = HttpMethod.Post;
            msg.RequestUri = new Uri("https://yourazuredureablefunction.azurewebsites.net/api/TextAnalysis_HttpStart?code=yourfunctionaccesscode");
            msg.Headers.Add("Ocp-Apim-Subscription-Key", "yourkeyfromcognitiveservices");
            msg.Headers.Add("Ocp-Apim-Subscription-Region", "yourresourceregion");
            msg.Headers.Add("Ocp-Apim-Subscription-Url", "https://api.cognitive.microsofttranslator.com/translate?api-version=3.0&to=en");
            msg.Headers.Add("Ocp-Apim-Subscription-Method", "Translation");
            
            object data = new
            {
                displayName = "Document Summarization Task Example",
                analysisInput = new
                {
                    documents = new[] { new { id = 1, language = "en", text = longtxt } }
                },
                tasks = new[]
                {
                    new
                    {
                        kind = "ExtractiveSummarization",
                        taskName = "Document Summarization Task",
                        parameters = new { sentenceCount = 3 }
                    }
                }
            };

            StringContent content = new StringContent(JsonConvert.SerializeObject(data), Encoding.UTF8, "application/json");
            content.Headers.ContentType = new MediaTypeHeaderValue("application/json");
            msg.Content = content;

            Stopwatch sw = new Stopwatch();
            sw.Start();
            HttpResponseMessage response = _httpClient.Send(msg);
            string resp = response.Content.ReadAsStringAsync().GetAwaiter().GetResult();
            dynamic durableDyn = JsonConvert.DeserializeObject<dynamic>(resp);

            string statusUrl = durableDyn["statusQueryGetUri"];

            HttpRequestMessage statusMsg = new HttpRequestMessage();
            statusMsg.Method = HttpMethod.Get;
            statusMsg.RequestUri = new Uri(statusUrl);

            HttpResponseMessage statusResponse = _httpClient.Send(statusMsg);
            string statusResp = statusResponse.Content.ReadAsStringAsync().GetAwaiter().GetResult();
            dynamic statusDyn = JsonConvert.DeserializeObject<dynamic>(statusResp);

            while (statusDyn["runtimeStatus"] != "Completed")
            {
                Thread.Sleep(TimeSpan.FromMilliseconds(1000));

                statusMsg = new HttpRequestMessage();
                statusMsg.Method = HttpMethod.Get;
                statusMsg.RequestUri = new Uri(statusUrl);

                statusResponse = _httpClient.Send(statusMsg);
                statusResp = statusResponse.Content.ReadAsStringAsync().GetAwaiter().GetResult();
                statusDyn = JsonConvert.DeserializeObject<dynamic>(statusResp);
            }

            sw.Stop();
            Console.WriteLine($"Time taken: {sw.ElapsedMilliseconds} Milliseconds");
            JArray outputs = statusDyn["output"];
            foreach (JValue output in outputs)
            {
                Console.WriteLine(output.ToString());
            }
        }
    }
}

Finally, a sample code for Translation Language Detection service, which in theory works for detecting multiple languages.

using Newtonsoft.Json;
using Newtonsoft.Json.Linq;
using System.Diagnostics;
using System.Net.Http.Headers;
using System.Text;

namespace Tester
{
    internal class Program
    {
        static HttpClient _httpClient = new HttpClient();
        static void Main(string[] args)
        {
            string longtxt = File.ReadAllText("news_text.txt");

            HttpRequestMessage msg = new HttpRequestMessage();
            msg.Method = HttpMethod.Post;
            msg.RequestUri = new Uri("https://yourazuredureablefunction.azurewebsites.net/api/TextAnalysis_HttpStart?code=yourfunctionaccesscode");
            msg.Headers.Add("Ocp-Apim-Subscription-Key", "yourkeyfromcognitiveservices");
            msg.Headers.Add("Ocp-Apim-Subscription-Region", "yourresourceregion");
            msg.Headers.Add("Ocp-Apim-Subscription-Url", "https://api.cognitive.microsofttranslator.com/detect?api-version=3.0");
            msg.Headers.Add("Ocp-Apim-Subscription-Method", "TranslationForLanguageDetection");
            
            object[] data = new object[] { new { Text = longtxt } };

            StringContent content = new StringContent(JsonConvert.SerializeObject(data), Encoding.UTF8, "application/json");
            content.Headers.ContentType = new MediaTypeHeaderValue("application/json");
            msg.Content = content;

            Stopwatch sw = new Stopwatch();
            sw.Start();
            HttpResponseMessage response = _httpClient.Send(msg);
            string resp = response.Content.ReadAsStringAsync().GetAwaiter().GetResult();
            dynamic durableDyn = JsonConvert.DeserializeObject<dynamic>(resp);

            string statusUrl = durableDyn["statusQueryGetUri"];

            HttpRequestMessage statusMsg = new HttpRequestMessage();
            statusMsg.Method = HttpMethod.Get;
            statusMsg.RequestUri = new Uri(statusUrl);

            HttpResponseMessage statusResponse = _httpClient.Send(statusMsg);
            string statusResp = statusResponse.Content.ReadAsStringAsync().GetAwaiter().GetResult();
            dynamic statusDyn = JsonConvert.DeserializeObject<dynamic>(statusResp);

            while (statusDyn["runtimeStatus"] != "Completed")
            {
                Thread.Sleep(TimeSpan.FromMilliseconds(1000));

                statusMsg = new HttpRequestMessage();
                statusMsg.Method = HttpMethod.Get;
                statusMsg.RequestUri = new Uri(statusUrl);

                statusResponse = _httpClient.Send(statusMsg);
                statusResp = statusResponse.Content.ReadAsStringAsync().GetAwaiter().GetResult();
                statusDyn = JsonConvert.DeserializeObject<dynamic>(statusResp);
            }

            sw.Stop();
            Console.WriteLine($"Time taken: {sw.ElapsedMilliseconds} Milliseconds");
            JArray outputs = statusDyn["output"];
            foreach (JValue output in outputs)
            {
                Console.WriteLine(output.ToString());
            }
        }
    }
}

Services Supported

At the moment, the following services are supported, async operation is tested for 2022-10-01-preview version

Service	Sync Operation	Async Operation
Language Detect	X
Key Phrase Extraction	X	X
Entity Extraction	X	X
PII Redaction	X	X
Summarization		X
Entity Linking	X	X
Translation	X
Translation for Language Detection	X

License

Free software, absoltely no warranty, use at your own risk!