Keyword Tool
The Keyword Tool helps you to methodically create a list of relevant keywords and phrases describing your topic that can subsequently be searched for in earnings calls.
The Keyword Tool has two main purposes. First, by suggesting similar keywords as they are used in earnings calls, it helps to include keywords that may be relevant for the topic. Second, by asking the user to accept or reject a suggested keyword, it helps to exclude keywords that are irrelevant for the topic.
Example: Suppose the topic of interest is the Covid-19 pandemic. Merely searching for the keyword 'corona' may miss similar discussions that do not contain the keyword, such as 'COVID', 'coronavirus', or 'pandemic'. With the Keyword Tool all such relevant words can be found.
The Keyword Tool is trained on over a billion phrases in corporate earning calls to suggest keywords that are semantically similar to the input phases. The algorithm is based on FastText, an open-source, free, lightweight library developed by Facebook's AI Research (FAIR) lab. We extend the algorithm to support multi-word phrases.
How to use it
Using the Keyword Tool consists of two main steps:
Enter one or more keywords.
For example, when the user types in 'ChatGPT', the following results are depicted:
If several keywords are used, they should be separated with a comma so the tool can identify synonyms for each keyword.
Plurals are not automatically included in the search and may need to be added by the user if necessary.
Example: When searching for similar terms for a noun, say 'investment', the Keyword Tool will not assume the input implies plural too. To include the plural form, please enter both 'investment' and 'investments'.
Accept or reject suggested keywords.
Using the icons next to each keyword, the user can now accept or reject the suggested keywords according to the topic the user has in mind. Accepting a keyword means that it joins the list of accepted keywords. The accepted keywords are later to be used in the Risk Tool. Similarly, rejecting a keyword means that it will not join the list of accepted keywords.
For keywords whose relevance is not immediately clear, the Keyword Tool also helps the user to methodically decide about their relevance. The idea is to count the number of "correct" senses that the keyword is used based on a randomly-drawn set of sentences containing the keyword. If a high percentage of the sentences containing the keyword, say 28/30 (93.3%) sentences, use the keyword in the sense the user intends, one would accept the keyword.
To facilitate this process, the Keyword Tool has the book icon next to a keyword suggestion. When a user is unsure about a keyword's relevance, the user can click on that icon. The Keyword Tool will then show randomly-selected sentences containing that keyword together with two buttons ("Correct sense" and "Incorrect sense"). These two buttons can be used to keep track of the number of time the keyword was used in the intended sense.
Finally, clicking the "Save search protocol" button will generate a txt file with all information from the Keyword Tool: The accepted and rejected keywords, as well as the number of correct senses for each keyword that the user read randomly-selected sentences about.
Settings: your options for refining the search
Length of similar keyword:
This option allows the user to determine the length of the similar keywords that the Keyword Tool should find. The tool provides one-word responses, two-word responses, three-word responses, or any combination of those options. Please note that to use the 'Any length' option, the other options should be deselected first.
Select model:
The Keyword Tool has been trained for finding similar keywords on earning calls from three different time periods:
The first period lasts from 2020-2022 and is better able to predict synonyms for topics that have been discussed recently such as the Covid-19 pandemic, the Russian invasion of Ukraine, remote work, semiconductor shortage, etc.
The second time period includes all earning calls from the last ten years (2012-2022). It will have more data for topics since 2012 such as Fukushima, cloud services, streaming, subscription models, ordering online, social networks, cryptocurrencies, etc.
The third time period includes all earnings calls from the year 2023.
Number of keywords to return:
This option allows the user to control how many suggestions are returned with each search.
Capitalization:
By default, the Keyword Tool takes capitalization into account. This means it respects the difference between words like Turkey (country) vs. turkey (bird) or Trump (person) vs. trump (verb).
Last updated