Predictive Text Classification for a AdTech Company
The company mainly worked with musicians who sell their big data services. They have already collected data from basic websites and used it to show ads to visitors based on their preferences.
The client approached us with a business need to broaden the focus, analyze more complicated websites, and search for various types of information (to improve the accuracy of advertising). The idea was to create a model that could analyze millions of web pages per day and accurately categorize the content based on the valuable context results.
Inoxoft has supported the client’s intentions and delivered an ML model to perform categorization. Every word was given an IAB category in the model that could analyze websites and predict consumers’ preferences further. The model was developed as a multilingual classifier, allowing keyword websites to be written in any other language.
The first-class machine learning model successfully sells the agency's services by providing value and performance for the world’s largest brands and global media agencies.
- IAB categorizationÂ
- Custom Python Scraping BackendÂ
- Custom Tensorflow training and model deployment
Full Story
Inoxoft Solution
The company aims to transform the way organizations connect with their target audiences, empower businesses with actionable insights and high-quality data, by leveraging advanced technology and data science.
Covering the need to expand the focus and analyze complex websites, where one can find lots of different informational categories (the categories are eligible to extract valuable textual data about website visitors and allow serve advertising more accurately), Inoxoft, together with our client, created an ML model, which could analyze millions of these web pages per day and categorize the content precisely following the valuable context results.Â
To cover a bigger number of websites with extensive and different content, it was decided to perform natural language processing.Â
The content on any website was broken down into parts with the help of - tokenization, which split the text into separate elements - stop words detection, which helped delete all the useless commonly used words - lemmatization, which treated all the word forms as a single word - decontraction, which allowed a transition from shortened forms of words into full ones - text to sequences, which is a method of Tensorflow tokenizer that labeled the words and assigned them numbers - padding, which signed zero to any missing word - label encoding, which gave out a numbered listÂ
The team has used the IAB categories and machine learning modeling to perform text classification. The model allows us to understand what users like on any website and serve the right ads to the right people.
Results
Inoxoft's team has produced the first-class ML model for the client, allowing them to
- Analyze complex websites with different content
- Process 2 million websites per day - obtain the inference in under 10 milliseconds
- Get precise categorization of words and context
- Successfully sell the agency's services by promoting the right ads to the right audiences
