refactor(page_analyzer): use screenshot when inferring similar items

Author: idiotWu

npiai/tools/scrapers/page_analyzer/app.py

@@ -607,15 +607,16 @@ async def infer_similar_items_selector(
         # use latest page url in case of redirections
         page_url = await self.get_page_url()
         page_title = await self.get_page_title()
-        # raw_screenshot = await self.get_screenshot(full_page=True)
+        raw_screenshot = await self.get_screenshot(full_page=True)
 
         contentful_elements = await self.playwright.page.evaluate(
             """
-            () => npi.getMostContentfulElements(null, 10)
+            (screenshot) => npi.getMostContentfulElements(screenshot, 10)
             """,
+            raw_screenshot,
         )
 
-        # annotated_screenshot = await self.get_screenshot(full_page=True)
+        annotated_screenshot = await self.get_screenshot(full_page=True)
 
         elements_as_markdown = []
         group_element_count = {}
@@ -669,7 +670,7 @@ async def infer_similar_items_selector(
                         
                         ## Instructions
                         
-                        Follow the instructions to determine whether there is a pagination button on the current page for navigating to the next page:
+                        Follow the instructions to determine whether there are similar elements representing the most meaningful list of items:
                         1. Examine the URL, and the title of the page to understand the context, and then think about what the current page is.
                         2. Go through the elements array, grab the semantic information of the elements via the "content" property. Pay attention to the elements with the same group ID as they are under the same parent element.
                         3. Check if there are similar elements representing **the most meaningful list** of items. Typically, these elements link to the detail pages of the items. Note that these elements should not be the pagination buttons and should contain enough meaningful information, not just some short phrases.
@@ -691,12 +692,12 @@ async def infer_similar_items_selector(
                                 ensure_ascii=False,
                             ),
                         },
-                        # {
-                        #     "type": "image_url",
-                        #     "image_url": {
-                        #         "url": annotated_screenshot,
-                        #     },
-                        # },
+                        {
+                            "type": "image_url",
+                            "image_url": {
+                                "url": annotated_screenshot,
+                            },
+                        },
                     ],
                 ),
             ],