3 ms·

I tried with the query “principal software engineer” Keywords were not helpful and included company names For example: One of the keywords was “fidelity” An

by pjm331 3y ago

I tried with the query “principal software engineer” Keywords were not helpful and included company names For example: One of the keywords was “fidelity” Another keyword was “using” A third keyword was “disability” As far as requirements - some of the items on the list were actually requirements but others were just descriptions of specific companies and some were fragments of sentences that made no sense out of context For example: > This is a great time to join us and be part of this journey!Certifications:Company OverviewFidelity Investments is a privately held company with a mission to strengthen the financial well-being of our clients > It also guarantees that your data security strategy adheres to compliance and data privacy regulations Finally the list of requirements was very long - I’d expect the 10,000 foot view to be summarized a lot more - this seemed like it might be the same as skimming the 22 job descriptions it was generated from So imo you’ve got some work to do before this is ready for prime time

todyWasAGoodDy 3y ago
First of all thank you for taking the time and sharing your experience. > 10,000 foot view to be summarized a lot more This is so spot on and point taken. The idea at the results page was to have filters to allow users reduce the amount of results shown. But, yes you are right the requirements list returns `one mighty scroll`. I will have a think of ways to tame it. > fidelity I had mixed results for company recognition from a NER perspective, and this keyword I am afraid is the product of that. I have some ways to reduce reliance on NER here, will see if that fixes it. > disability these type of words are slightly more nuanced as in some cases these will be relevant keywords... this I will have to iterate more. > sentence samples I expect the model to get better over time to weed these out. I am somewhat weirdly confident about it as I have seen 1st hand how it missed obvious things in earlier versions, but is doing much better now. all in all thank you once again for taking the time, have a great day.

by pjm331 3y ago

todyWasAGoodDy 3y ago

First of all thank you for taking the time and sharing your experience. > 10,000 foot view to be summarized a lot more This is so spot on and point taken. The idea at the results page was to have filters to allow users reduce the amount of results shown. But, yes you are right the requirements list returns `one mighty scroll`. I will have a think of ways to tame it. > fidelity I had mixed results for company recognition from a NER perspective, and this keyword I am afraid is the product of that. I have some ways to reduce reliance on NER here, will see if that fixes it. > disability these type of words are slightly more nuanced as in some cases these will be relevant keywords... this I will have to iterate more. > sentence samples I expect the model to get better over time to weed these out. I am somewhat weirdly confident about it as I have seen 1st hand how it missed obvious things in earlier versions, but is doing much better now. all in all thank you once again for taking the time, have a great day.