FoundConf Elias Dabbas.pdf
本地来源：seo-llm/raw/seo知识库/TXT整理/seo方法论/ppt/John喜欢和引用的那些海外演讲嘉宾PPTs/FoundConf-Elias Dabbas.pdf.txt
Getting Started with Data Science and LLMs for Modern SEO Elias Dabbas October 28, 2025 Tokyo Me What? • Image: Drew Conway, remade with ChatGPT What? • Domain Expertise (SEO, finance, geology, etc): This drives the other elements by providing answers to “what” and “why” questions • Programming: basic programming skills (not software development), questions of “how” • Math and Statistics: average, maximum, minimum, modeling, machine learning, deep learning, etc. What? Tasks and skills • Image: R for Data Science book What? Tasks and skills • Import: read CSV file, crawl a website, get API data • Tidy: reshape a table, merge tables • Transform: sort, groupby (pivot tables), filter • Visualize: charts, interactive apps • Model: linear model, inference, clustering • Communicate: reports, dashboards, recommendations • Program: perform all tasks by writing code How? How? Writing (Python) code in a “Jupyter Notebook” Minimum viable knowledge Variables name = value • B2 = 10+5 • name = value (potentially a complex formula) •= is the “assignment operator” Functions =FUNCTIONNAME(parameter_1, parameter_2, parameter_3, …) Right-click (contextual functions) Dot-notation • Available actions depend on the type of object Jupyter Notebook File format Supporting applications File format Supporting applications JupyterLab .ipynb Interactive Python notebook Jupyter Notebook Google Colab https://colab.research.google.com/ Start here (online, no setup required) Kaggle Examples Crawling a website One line of code to crawl an entire website • Many more columns! Fetching/analyzing XML sitemaps Compare sitemap to crawled URLs Compare sitemap to crawled URLs 857 URLs 237 URLs Text Analysis • adv.word_frequency: ngram analysis of a list of phrases, optionally weighted by a metric URL structure analysis adv.url_to_df(openai[“url”]) adviz.url_structure() adviz.url_structure() Drill-down to the next level (folder) Why? Why work with code? vs. moving your mouse and clicking on the screen We need to trust that the work we do is correct Code to compare sitemap vs crawled URLs • Check every step of the process, verify things are correct, ask questions Horizontal: reusing the same workflow across projects Run the exact same process for another domain Vertical: improving the same workflow to make it more powerful Analyze links Get a mapping of all links on the site (from/to) Analyze links Top linked external domains (de-duplicated) Raw data + code = results Should be the same every time (like scientific experiments) Full customization A Great Tool Provides great charts Uncovers insights Makes interesting reports Allows you to ask any question & perform any task you want with your data Automation & packaging Let’s make that whole workflow into a single function Create a single function crawled_not_in_sitemap(“example.com", “output.jsonl”) a single command that goes through all the steps A set of functions that you can use without having to worry about their code do_this(a) do_that(b, c) Introducing advertools A Python package that has various functions and tools dedicated to SEO, and digital marketing open-source and free Getting started with advertools • Installation: pip install advertools • Importing: import advertools as adv • Running a function: adv.function_name(param_a=val_a, param_b=val_b) Robots.txt files • adv.robotstxt_to_df(): convert a robots file (or a thousand) to a DataFrame • adv.robotstxt_test(): run tests in bulk for a combination of URLs and user-agents XML Sitemaps • adv.sitemap_to_df(): convert a sitemap to a DataFrame • Supports normal sitemaps and sitemap index files • Supports robots.txt files (discovers all listed sitemaps and combines them together) • Runs concurrently (very fast) • Speed can be configured easily Crawling Three crawlers • adv.crawl(): normal SEO crawler • Spider and/or list mode • Get all request/response headers • Set granular rules on speed • Much more • adv.crawl_headers(): send HEAD requests in bulk to a set of known URLs, get status codes and all available response headers • adv.crawl_images(): download all images from a list of URLs, optionally set limits for minimum width/height, and name regex Crawling advertools crawler extracted (groups of) columns • adv.crawl(): normal SEO crawler • Spider and/or list mode • Get all request/response headers • Set granular rules on speed URL Analysis • adv.url_to_df: split a list of URLs to their components and return a DataFrame Log file analysis • adv.logs_to_df: comprehensive parsing of log files • Parses every single field into its components (URLs, referer and request, useragent, status codes, etc) • Extremely granular analysis of log data • Automatic compression with parquet format adviz advertools vizualizations • adviz.serp_heatmap() The DataFrame Mentality © Ray Grieselhuber AI/LLMs and Data Science The dance between (un)structured tasks Image: ChatGPT The Cycle • Get unstructured data • Provide some structure/context with traditional methods • Extract structured data • Use traditional software for processing • Repeat Image: ChatGPT Crawled website’s body text Summarize in bulk with AI • Create an empty list summaries • Go through all the body_text column cells in the crawldf • Send the content to ChatGPT with a prompt asking it to summarize it • Append the summary text to summaries Crawled website’s body text 90% of My Skills Are Now Worth $0 ...but the other 10% are worth 1000x Kent Beck (software developer) https://x.com/KentBeck/status/1648413998025707520 90% of My Skills Are Now Worth $0 ...but the other 10% are worth 1000x 90% $0 10% $10,000 • Memorizing which package does what • Asking the right, strategic, useful questions • Knowing the syntax of the functions that you use • Making a good evaluation of the given output • Knowing what methods are available to Python dictionaries • Making decisions on next steps • Writing code • Telling the LLM to write code “AI doesn’t do it end-to-end. It does it middle-to-middle. The new bottlenecks are prompting and verifying.” Balaji Srinivasan (entrepreneur) https://x.com/balajis/status/1937517664907460980 Human AI/LLM Human End Middle End Prompting Verifying @EliasDabbas
本文档为站内渲染。原始文件本地路径：saas/source/seo-llm/raw-seo知识库-TXT整理-seo方法论-ppt-John喜欢和引用的那些海外演讲嘉宾PPTs-FoundConf-5eea0d.txt（仅本地保留，不入库不部署）