MY PITCH IS GOOD by Yves Curtat. Chairman and Founder of Retail Reload. In this exclusive interview, Yves explains the performance that his RFID solution with AI brings to Retail.
Using a multi-criteria algorithm with artificial intelligences like ChatGPT!
We give you some information in this article.
Technologies used to make decisions by taking several criteria into account simultaneously.
They can be used in many fields, including artificial intelligence, to help solve complex problems.
For example, if you want to train an AI model to make decisions based on several criteria. This algorithm allows you to evaluate and compare the different options available.
In the case of ChatGPT, the use of multi-criteria algorithms can be useful to improve the generation of responses by taking into account several objectives simultaneously.
For example, you might want the responses generated to be relevant, consistent and politically correct.
In addition, a multi-criteria algorithm can be used to evaluate and rank the responses generated according to these criteria, and thus select the most appropriate response to display.
However, it is important to note that the specific implementation of multi-criteria algorithms with ChatGPT will depend on the technical details and specific objectives you wish to achieve.
Common multi-criteria optimisation techniques include the use of weighting methods. But also methods based on dominance approaches. Or even methods based on mathematical programming approaches.
Request analysis: When you receive a request via a form, you can use ChatGPT to analyse the request and extract the keywords that make it up.
Additional information in the prompt with the request may be useful. This allows ChatGPT to better understand the context.
Database: You need to have a database of detailed information about the items you are looking for.
Each item, which we will call an “Item”, must be associated with relevant keywords corresponding to dedicated and unique elements relating to it.
Keyword comparison: Once you have extracted the keywords from the request and have access to the database of “Items”, you can use a multi-criteria algorithm to compare the keywords in the request with the keywords associated with the “Items”.
This can include criteria such as keyword match or other specific criteria that you can define.
The multi-criteria algorithm can assign a weighting to the different criteria according to their relative importance. For example, exact keyword matches may be considered more important.
The algorithm can then evaluate each Item according to these criteria and generate a list of ‘ideal’ Items that best match the search you have initiated.
Displaying results. You can display the Items that best match the query. The results can be sorted according to their relevance, using the scores or rankings calculated by the multi-criteria algorithm.
It is important to note that the specific implementation of this multi-criteria algorithm will depend on the specific needs and constraints of your application.
It is also important to bear in mind factors such as the confidentiality of user data and the protection of privacy when handling any personal information.
When searching for items in the database, it is advisable to use advanced search techniques to filter out items that include keywords or have approximate matches.
Exact match: You can use an exact match to filter out Items that include exactly the keywords. This ensures that the selected items correspond directly to the specified keywords.
However, this may be too restrictive and exclude relevant profiles that may use similar or synonymous terms.
Stemming and lemmatisation: used to standardise the keywords and words contained in the Items. Stemming consists of reducing words to their root form, while lemmatisation aims to obtain the lemma, i.e. the canonical form of a word. This allows you to search for variants of words that share a common root, which can help to broaden the correspondence between keywords and Items.
Fuzziness: allows you to take account of approximate matches. For example, the fuzzy matching algorithm measures the similarity between two strings by counting the minimum number of operations required (insertions, deletions or substitutions) to transform one into the other.
This can be useful for finding profiles with spelling variations or keyword typos.
Text similarity search: You can also use text similarity search techniques, such as vector search or reverse indexing, to find Items that are similar to the keywords in your query.
These techniques use similarity measures to compare the representation vectors of keywords and Items and return those that are most similar.
It’s important to note that you can combine these techniques and adapt them to suit your specific needs.
For example, you can use exact matching for important keywords, while using stemming, lemmatisation and fuzzy matching techniques for less strict keywords.
ChatGPT can help you to expand your keyword list to get a better idea of your query, rather than just the keywords. This can be particularly useful when your query is vague or incomplete.
Gathering information: When you submit a query to ChatGPT, you can provide descriptive information. This information will help ChatGPT to better understand the context.
Semantic analysis: Using the information provided, ChatGPT can perform a semantic analysis to extract the relevant terms that describe the object in relation to the query.
This creates richer, more representative lists that you can use for searching.
Rather than focusing solely on specific keywords, you can search for matches between terms in these lists from your database and broaden the scope of the search.
It’s important to note that this is a complex and constantly evolving area. Not least with the use of pre-trained language models like ChatGPT that can be ‘finetuned’ for specific tasks.
The use of multi-criteria algorithms with artificial intelligence such as ChatGPT can be beneficial for improving decision-making and generating responses by taking into account several criteria simultaneously.
By using ChatGPT to perform an analysis of the information provided in the query.
You can obtain additional information to better understand the query and search for matches with items in your database, enabling you to find relevant items even if your query is vague or incomplete.
However, a specific approach will depend on your application’s objectives and constraints
Photo credit : unsplash Mariia Shalabaieva & Mojahid Mottakin
–> Back to News
Consulting Tech Lead at GOWeeZ
With 15 years' experience in digital marketing, Christophe is familiar with managing complex digital marketing and innovation projects. He builds custom marketing tools (Design pattern MVC under Fat-Free Framework, etc.). POC and MVP development for GOWeeZ.
Christophe Rolland Tweet