The modern-age fashion industry has revolutionized fashion perception and amplification, but not for good. Carbon emissions, chemical pollution, greenhouse emissions, landfill waste, and dangerous working conditions are just some examples of the serious consequences we have to deal with today.
- The ethical fashion market has reached $7,548 million in 2022
- Over time, the segment is expected to grow at a compound average growth rate of 8,6% from 2027 and outmatch $16,819 million by the year 2023
There are several factors that make apparel manufacturers switch strategies and embrace eco-friendly fashion. These comprise consumer awareness and interest, social media, governmental initiatives and regulations, evolving methodologies, and more.
Yet still, there are many unyielding clothing manufacturers that refuse moving towards sustainable business. Some factors that restrain change include high costs, low margins, limited material, lacking standardization.
Our project: Quick overview
In 2022, our company was approached by a promising startup that promotes sustainable lifestyle and fashion. In brief, our engineers were contracted to investigate the opportunity of building a marketplace for consumers to search relevant information about whether individual brands are implementing sustainable policies.
Our team has covered:
- The composure of a PoC product to confirm the feasibility of the planned project
- The development of an NLP model
By utilizing machine learning, we introduced advanced search:
- First, the NLP system helps identify and extract relevant details from different document elements — headers, footers, tables, charts, and others
- Then, the AI solution automatically categorizes and scores pertinent details by gathering the answers to specific questions, both closed and opened
Mainly prioritizing user experience, we ensured easy navigation:
- Step 1: The user must select the brand he wants to know more about
- Step 2: The user must choose the metric (for example, the percentage of utilized renewable energy)
- Step 3: Keyword specification — the user can add relevant keywords to make the results more precise
- Step 4: Information scraping — the user can scrap gathered information to review article descriptions, add or remove links
- Step 5: At the next step, the user can create a report to display the results in a comprehensive table
- Step: At the last step, the user can download the report to save the results
The selection and training of the NLP model: Our contribution
Having extensive technical expertise in implementing artificial intelligence, we covered in-depth investigation to confirm the feasibility of the planned project.
First thing, our team has researched and selected a suitable NLP model to meet the client’s business target. After that, our engineers have prepared and trained the appropriate NLP model to gather relevant information from different heterogeneous sources.
The solution now processes received data in the following steps:
- Data collection
- Data scraping
- Data validation
- Data evaluation and scoring
By leveraging modern technology, we delivered an unprecedented NLP marketplace to enable potential users to source up-to-date information about brands and their sustainability policies.
To tailor the designed AI solution to the pre-defined requirements, we went through multiple iterative stages. This means, the delivered ML model was compiled from several existing models to process different inquiries.
The solution can cover:
- Basic questions with a specific keyword
- General questions about a specific topic
- Open questions, which require exact answers
- Close questions, which require exact answers
There were two challenges directly associated with the NLP model:
- Model selection
The implemented NLP solution had to be capable of handling different formulations of the same information. To resolve this challenge, we combined multiple models to handle ambiguous documents.
- Model training
The designed NLP solution had to be capable of processing complicated requests, including closed and opened. To overcome potential failure, we simplified possible requests to obtain an appropriate training basis.
Another challenge was associated with obtaining comprehensive information about brands:
- Some companies aren’t publishing trustworthy information about implemented sustainability policies. This means, they’re not open about production and labor conditions, materials used, waste disposal, and other important details.
- Some companies don’t own production premises –outsourcing fabrication has become quite popular. Many businesses delegate manufacturing to India, China, Turkey, and other far-located countries, which makes obtaining information about their production conditions either challenging or even completely impossible.
- The accessibility of information might depend on the structural organization within companies.
- Some brands conceal information to avoid reputational damage.
By leveraging domain-specific knowledge and experience in providing full-cycle custom software development, we delivered a marketplace, which helps easily identify ethical brands that prioritize sustainable production.
After resolving the challenges, which often come with advanced technology, in particular artificial intelligence, we provided the client with a cutting-edge solution, which encompasses data collection, scraping, validation, automatic validation, and scoring.
At the first stage of the AI development, we covered:
- Research automation
- Source clearing
- Scraper implementation to extract relevant information from both HTML and PDF files
- Data transformation
During the next steps of the ML development, we handled:
- Primary research process validation — questions about one brand
- Secondary research process validation — questions about multiple brands
- The extension of the NLP model
- The integration of the NLP model