A few days ago, news about the "noodle refill" incident in Jiajiang County, Leshan, Sichuan Province, became a hot search topic, and someone also shared this news in the "Malá Là GIS" group chat. While debates about cultural differences and other issues were still raging on Weibo, a breath of fresh air appeared in our GIS group when someone asked: How can we obtain POI data for restaurants that offer "free noodle refills"? (See? Our GIS group truly has a professional spirit.)
As you know, I usually don't slack off, but I couldn't resist such an interesting GIS question. I consulted with various experts and did some hands-on experimenting. Although the final result wasn't entirely satisfactory, I'll still summarize the process for everyone today.
Traditional POI Databases
I've shared many POI datasets before, the most famous being the POI dataset from OSM. However, I found that the data fields provided by such sources are currently too simplistic, as shown below:
There's only a simple name and coordinates. Figuring out whether they support free refills is impossible. What about Amap (Gaode Map)? The fields in Amap's POI data seem richer, but I still couldn't find detailed fields like "noodle refill".
Food Delivery Apps
Where would such detailed fields exist? Of course, we must look at food delivery apps. I searched on JD Instant Delivery, Taobao Flash Delivery, and Meituan, and found accurate results on all. Among them, Meituan, having acquired Dianping and offering in-store discounts, has more comprehensive data.
So, the next task is to scrape data from these food delivery apps.
Fiddler Packet Capture Practice (Unsuccessful Version)
Since the data is within the apps, obtaining it requires setting up packet capture. I used the Fiddler tool. Official website:
https://www.telerik.com/fiddler
Configuring device packet capture was more troublesome with older versions, but the new version is almost zero-configuration; just follow the guide step by step. (It must be said, the user experience of this software is excellent.)
Opening this configuration leads to a setup wizard. Using Android as an example:
You can proceed directly as required. After configuration, connect the phone and computer to the same network, then add a proxy to the phone's Wi-Fi settings:
Then restart Fiddler and open the phone app to see the captured packet data.
However, I didn't find specific parsed data. After all, Meituan's anti-scraping measures are top-tier in the industry; it wouldn't be normal if it were so easily captured. I briefly checked; usually, modern apps use SSL Pinning technology. Common countermeasures involve Frida script injection; another method is to use Apktool for decompilation and repackaging with a new signature. I haven't done much of the above methods, tinkered a bit, and lacking the relevant devices, I gave up. Interested friends can research it further.
AI Assistance
Since app packet capture failed, in the past, this situation would mean either giving up or finding paid solutions. However, it's 2026 now, and AI is very powerful. Since the data volume isn't large, we can use AI to brute-force a solution.
First, take a long screenshot of the search results from the app, then upload the image to Doubao AI, paired with the following prompt:
The user's screenshot location is xxxxxxx. Please analyze the content of the screenshot, summarize all the stores in the image, and search for their detailed addresses.The returned result is as shown:
If the data seems correct, we can use the following prompt to obtain latitude and longitude coordinates:
Organize the above content into an Excel file and add the latitude and longitude of these stores to the table.The final result is as shown:
Summary
Anti-scraping measures in apps are constantly evolving. Much of my previous knowledge is outdated. Using Fiddler might still work for ordinary small apps, but dealing with apps from major companies requires more professional tools.
Furthermore, with the support of AI today, what previously seemed like very complex manual data organization tasks can be done faster and more securely using AI. Often, when encountering similar problems, consider using AI to change your approach; you might just find a path forward.