Licy: A Chatbot Assistant to Better Understand and Select Open Source
Software Licenses
Giorgos Shittas, Georgia M. Kapitsaki
a
and Maria Papoutsoglou
b
Keywords:
Open Source Software Licenses, Chatbot, License Recommendations.
Abstract:
Open Source Software (OSS) carries licenses that specify the terms under which the software is made available
for use. Various resources are available for software engineers online, in order to assist them in understand-
ing and choosing among the available OSS licenses when creating their software projects. However, these
resources lack in the provision of a sense of interactivity to user prompts, which would have been useful for
providing guidance in a more familiar manner. In this work, we present our approach for Licy, a chatbot OSS
licensing assistant for guiding users with information on specific OSS licenses and for choosing which OSS
licenses to use in specific cases. A large number of licenses are supported by the chatbot using the license
model offered by choosealicense, focusing on license permissions, limitations and conditions. We describe the
design and implementation process of the chatbot and its preliminary evaluation results using chatbot design
metrics and a user evaluation. We argue that the chatbot can serve as a starting point for similar interactive
assistants for software engineers, and describe its value in that respect.
1 INTRODUCTION
Open Source Software (OSS) is widely used, while
many software systems are made available as OSS,
even by large vendors. OSS licenses specify the
terms under which the original software is available
for use by the licensees, the users of the software, and
how it can be further modified and distributed. The
OSS licensing landscape is complex, as it contains a
wide variety of licenses that are linked with differ-
ent combinations of permissions, constraints, and re-
quirements (Laurent, 2004). A very large number of
OSS licenses exists nowadays, with Software Package
Data Exchange (SPDX)
1
listing 634 OSS and con-
tent licenses including their different versions (Stew-
art et al., 2010). Even among the more limited number
of popular licenses, understanding the license content
and the context of use is not trivial. The OSS li-
censes that are the most dominant today, according to
Statista,
2
are the GNU General Public License (GPL),
the Apache License, and the MIT License.].
Although there are available resources for OSS li-
a
https://orcid.org/0000-0003-3742-7123
b
https://orcid.org/0000-0003-0658-5065
1
https://spdx.org/licenses/
2
https://www.statista.com/statistics/1245643/
worldwide-leading-open-source-licenses/
censes online, e.g. TLDRLegal
3
and choosealicense
4
,
and prior research works that have focused on license
recommendations and guidance (Kapitsaki and Char-
alambous, 2019; Kapitsaki et al., 2022; Xu et al.,
2023), there is a lack of tools that are interactive when
it comes to understanding and selecting OSS licenses.
In this work, we use as starting point the above to
explore the potential of chatbots as a solution to the
issue of OSS licensing assistants. Through simulat-
ing human-like conversations, chatbots deliver instant
responses, and automate functions that once required
a human touch, improving the user experience (Lok-
man and Ameedeen, 2019). We are introducing the
design and implementation of an OSS licensing assis-
tant chatbot, Licy, that aims at simplifying concepts
found in OSS licenses and at giving users a conversa-
tional environment for finding the license that is most
appropriate for their software project. The chatbot
supports 58 licenses, and its prototype implementa-
tion has been evaluated using chatbot design metrics
from a prior publication (Ca
˜
nizares et al., 2022) and
via the participation of 18 users with limited experi-
ence in OSS. The results of this evaluation show that
Licy is straightforward to use, while it has the poten-
3
https://www.tldrlegal.com/
4
https://choosealicense.com/
Shittas, G., Kapitsaki, G. M. and Papoutsoglou, M.
Licy: A Chatbot Assistant to Better Understand and Select Open Source Software Licenses.
DOI: 10.5220/0013336400003928
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 20th International Conference on Evaluation of Novel Approaches to Software Engineering (ENASE 2025), pages 573-581
ISBN: 978-989-758-742-9; ISSN: 2184-4895
Proceedings Copyright © 2025 by SCITEPRESS – Science and Technology Publications, Lda.
573

tial to be expanded with additional features.
We argue that this work can serve as a starting
point for similar conversational agents in the area of
OSS licensing, especially for educational pusposes.
This interactive approach allows the chatbot to ask
specific questions, clarify users’ intentions, and give
recommendations, with which the user can interact in
order to get more information on specific responses
provided by Licy.
The remainder of the text is structured as follows.
Section 2 presents OSS licenses and related work in
the area. Section 3 is dedicated to the presentation of
the chatbot design and implementation. The evalua-
tion performed is detailed in section 4. Findings are
further discussed in section 5, and section 6 concludes
the work outlining directions of future work.
2 BACKGROUND AND RELATED
WORK
2.1 Open Source Software Licenses
The essence of software licensing is to protect the in-
tellectual property rights of developers, as well as to
specify the rights and restrictions given to licensees.
Software licenses range in general from strict propri-
etary to permissive OSS licenses, each tailored to in-
dividual goals. When using software that is made
available under an OSS license, the user has the
opportunity to add enhancements and distribute the
modifications made, usually under the same licens-
ing conditions as the original software. Understand-
ing and being compliant with OSS licenses is vital
in respecting the rights of software creators. It also
makes users of software conscious of any limitations
the software they are using has, like prohibitions to
use derivative works for commercial purposes or con-
ditions to share modifications under the same license.
Unlawful use of software resulting from the lack of
indicating any license or from the violation of licens-
ing terms can cause legal disputes, so proper selection
and use of OSS licenses is vital (Reddy, 2009). OSS
licenses are divided in the following main categories:
• Permissive Licenses: Licenses in this category
do not impose tight restrictions on how software
can be used, modified or distributed. Usually,
they require that certain conditions are met when
the software is redistributed, such as inclusion of
the original license notices, copyrights and trade-
marks. Examples of permissive licenses include
the Apache license, the BSD (Berkeley Source
Distribution) licenses and the MIT License.
• Weak Copyleft Licenses: Licenses in this cate-
gory allow users to choose any licensing model
for their own (added) code provided that the orig-
inal software is not modified. LGPL (Lesser
General Public License) licenses and MPL-2.0
(Mozilla Public License 2.0) are the most typical
examples of this category.
• Strong Copyleft Licenses: They guarantee that
any distributed version or modification of the li-
censed original software, or any project that uses
it, will be made available under the same licens-
ing terms. This means that any software product
that employs source code licensed under a strong
copyleft license also needs to carry a strong copy-
left license, with examples including GPL and
AGPL (GNU Affero General Public License).
2.2 OSS Licensing Assistants
Prior works have focused on assisting users in choos-
ing the most appropriate license for their software.
findOSSLicense is a recommender system that assists
developers in selecting an appropriate license for their
software project (Kapitsaki and Charalambous, 2019;
Kapitsaki et al., 2022). The users can indicate a num-
ber of requirements, resulting from their responses to
specific questions, including rights they want to pro-
vide to users of the software, other licenses they are
already using in libraries in their source code, and
technical properties of the software, whereas licenses
recommended to other users of the recommender sys-
tem are also considered, when making recommenda-
tions. License compatibilities are also taken into con-
sideration using FOSSology for scanning the source
code (Gobeille, 2008). LicenseRec is a recommenda-
tion tool that is based on compatibility checks on OSS
projects’ code and dependencies (Xu et al., 2023).
It uses a compatibility matrix for compatibilities be-
tween licenses, whereas the source code is scanned
via the Scancode
5
tool to find licenses in project files.
When it comes to license terms, OSLiFe-DiSC is a
tool that helps OSS developers in understanding open
source licenses terms (Sassi, 2024). OSLiFe-DiSC
provides users the list of OSS licenses approved by
the Open Source Initiative (OSI),
6
it allows users to
identify which licenses support a specific set of fea-
tures, while it can make comparisons between two li-
censes. At the same time, online resources provide
more information on the terms of licenses. The afore-
mentioned TLDRLegal is based on legal analysis of
licenses and presents what the licensee can, cannot
5
https://github.com/aboutcode-org/scancode-toolkit
6
https://opensource.org/
ENASE 2025 - 20th International Conference on Evaluation of Novel Approaches to Software Engineering
574

and must do with each license, while choosealicense
offered by GitHub provides a basic set of licenses the
user can choose from based on his/her needs and also
lists the main terms of a large number of licenses.
2.3 Chatbot Applications
Chatbots have been used in various areas, including
customer service, healthcare and education. In
customer service, chatbots are employed to respond
to queries, solve problems, and lead users through
troubleshooting, such as in the case of social media,
where a chatbot trained using Twitter conversations
between users and agents was created (Xu et al.,
2017). Within healthcare, chatbots help with ap-
pointments, schedules and patient education (Shinde
et al., 2021). In education, chatbots can act as virtual
tutors and learning assistants, providing students with
interactive learning environments (Clarizia et al.,
2018). Chatbots have also been useful for scientific
applications, such as for querying multidimensional
big data (Franciscatto et al., 2022), and for software
engineering (Lebeuf et al., 2017), whereas the
wide spread of Large Language Models (LLMs)
have offered new directions for integration with
chatbots (Dam et al., 2024). In the area of software
licenses, Microsoft License Advisor
7
is a custom
version of GPT-4 that assists users with advice on
Microsoft licenses.
Relation to Previous Works. Existing online re-
sources and platforms on OSS licensing offer a static
user experience, missing the interactive aspect, which
is available with a chatbot interface targeted in the
framework of the current work. The conversational
nature and variety of licenses supported in the cur-
rent work make the chatbot an enhancement upon
static approaches to OSS licensing assistance, with
Licy building upon the existing static resources, e.g.
the licensing terms of choosealicense have been used
for license modeling purposes.
3 DESIGN AND
IMPLEMENTATION
3.1 License Modeling
The chatbot supports the licenses listed in Table 1.
The abbreviation of the license name follows the nam-
ing convention of SPDX (Stewart et al., 2010) that
7
https://chatgpt.com/g/
g-p7NpdszNq-ms-license-advisor
has been utilized in prior works on OSS licensing and
are used this way in the text (Kapitsaki and Kramer,
2014). Although not intended for software licensing,
we have also added the following Creative Commons
licenses due to their popularity, as they are also li-
censes that gather the attention of developers and are
listed by SPDX: CC0-1.0, CC-BY-4.0, CC-BY-NC,
CC-BY-ND-4.0, CC-BY-SA-4.0, CC-NC-ND, CC-
NC-SA, CC-ND, CC-SA. For the same reasons, we
have also added the content licenses GFDL-1.3 (GNU
Free Documentation License v1.3) and OFL-1.1 (SIL
Open Font License 1.1) (Lin et al., 2009).
For each license, we adopted the license terms
available in choosealicense in order to be able to
present to the user what the licensee is allowed to do
(can or permissions), what he/she is not allowed to do
(cannot or limitations) and which are his/her obliga-
tions (must or conditions). These terms are also used
when recommending licenses to users. An example
for MPL-2.0 license as provided by choosealicense is
shown in Table 2.
3.2 Overall Chatbot Workflow
We have chosen to design Licy as a conversation-
centric chatbot that mimics human discussions, as
this type of chatbots are considered better in natu-
ral conversations (compared to other types of chat-
bots) (Moore and Arar, 2018). The functionality of
the chatbot concerning licensing has as goals to:
1. Simplify the license selection process for users by
recommending appropriate licenses.
2. Provide explanations on OSS licenses and answer
relevant users’ questions.
The workflow for the user was designed to cover
the two above areas, as depicted in Figure 1, that in-
cludes also the Rasa
8
process stack
9
(at the bottom part of
figure) that is the framework used for Licy’s implementa-
tion. The Rasa framework has been chosen due to its pop-
ularity and since it has an Open Source edition (Ca
˜
nizares
et al., 2022). After the user provides his/her input, the in-
tention checker is used by the chatbot to analyze the user’s
input and decide on the actual intention. This is an impor-
tant stage where the chatbot determines what the users in-
tend to achieve with their query. When it comes to recom-
mending licenses, the chatbot is intended for beginner (Be-
ginner Questionnaire in the figure) or advanced users (Ad-
vanced Questionnaire in the figure), where beginners users
are shown a more limited set of questions. If license rec-
ommendation is detected as the user’s intent, these relevant
questions are shown to the user (the user indicates whether
he/she will be treated as a beginner or an advanced user).
8
https://rasa.com/
9
https://kavindasenarathne94.medium.com/
rasa-architecture-for-clever-chatbots-4a36b0b0ffcc
Licy: A Chatbot Assistant to Better Understand and Select Open Source Software Licenses
575

Table 1: OSS licenses covered by Licy.
License abbreviation and full name
1. 0BSD (BSD Zero Clause License)
2. AFL-3.0 (Academic Free License v3.0)
3. AGPL-3.0 (GNU Affero General Public License v3.0)
4. AML (Apple MIT License)
5. Apache-2.0 (Apache License 2.0)
6. APL-1.0 (Adaptive Public License 1.0)
7. APSL-2.0 (Apple Public Source License 2.0)
8. Artistic-2.0 (Artistic License 2.0)
9. BSD-2-Clause (BSD 2-Clause ”Simplified” License)
10. BSD-3-Clause (BSD 3-Clause ”New” or ”Revised” Li-
cense)
11. BSD-4-Clause (BSD 4-Clause ”Original” or ”Old” Li-
cense)
12. BSL-1.0 (Boost Software License 1.0)
13. CATOSL-1.1 (Computer Associates Trusted Open
Source License 1.1)
14. CDDL-1.0 (Common Development and Distribution
License 1.0)
15. CECILL-2.1 (CeCILL Free Software License Agree-
ment v2.1)
16. CPL-1.0 (Common Public License 1.0)
17. CUA-OPL-1.0 (CUA Office Public License v1.0)
18. Cube (Cube License)
19. ECL-2.0 (Educational Community License v2.0)
20. EFL-1.0 (Eiffel Forum License v1.0)
21. EFL-2.0 (Eiffel Forum License v2.0)
22. EPL-1.0 (Eclipse Public License 1.0)
23. ErlPL-1.1 (Erlang Public License v1.1)
24. EUDatagrid (EU DataGrid Software License)
25. EUPL-1.1 (European Union Public License 1.1)
26. EUPL-1.2 (European Union Public License 1.2)
27. GPL-2.0 (GNU General Public License v2.0 only)
28. GPL-3.0 (GNU General Public License v3.0 only)
29. ISC (ISC License)
30. LGPL-2.1 (GNU Lesser General Public License v2.1
only)
31. LGPL-3.0 (GNU Lesser General Public License v3.0
only)
32. LPPL-1.3c (LaTeX Project Public License v1.3c)
33. MIT (MIT License)
34. MIT-0 (MIT No Attribution)
35. MPL-2.0 (Mozilla Public License 2.0)
36. MS-PL (Microsoft Public License)
37. MS-RL (Microsoft Reciprocal License)
38. MulanPSL-2.0 (Mulan Permissive Software License,
Version 2)
39. NCSA (University of Illinois/NCSA Open Source Li-
cense)
40. ODbL-1.0 (Open Data Commons Open Database Li-
cense v1.0)
41. OSL-3.0 (Open Software License 3.0)
42. PostgreSQL (PostgreSQL License)
43. Unlicense (The Unlicense)
44. UPL-1.0 (Universal Permissive License v1.0)
45. Vim (Vim License)
46. WTFPL (Do What The F*ck You Want To Public Li-
cense)
47. Zlib (zlib License)
The questions shown to beginner users have been adopted
from the main choices offered by choosealicense. We chose
this option in order to keep the available choices for begin-
ner users limited and simple.
Table 2: MPL-2.0 license modeling based on chooseali-
cense terms.
Key Value
title Mozilla Public License 2.0
spdx-id MPL-2.0
description Permissions of this weak copyleft license are
conditioned on making available source code
of licensed files and modifications of those
files under the same license (or in certain
cases, one of the GNU licenses). Copyright
and license notices must be preserved. Con-
tributors provide an express grant of patent
rights. However, a larger work using the li-
censed work may be distributed under differ-
ent terms and without source code for files
added in the larger work.
permissions commercial-use
modifications
distribution
patent-use
private-use
sublicense
conditions disclose-source
include-copyright
same-license
limitations liability
trademark-use
warranty
The questions shown to advanced users are more: 11
in total but based on the reply of the user to a given ques-
tion, some questions might be skipped. In order to define
these questions, we relied on prior works on recommend-
ing licenses, that we adapted considering the expanded list
of licenses covered in the current work in relevance to fin-
dOSSLicense (Kapitsaki and Charalambous, 2016; Kapit-
saki and Charalambous, 2019). The questions added for
advanced users are listed in Table 3, whereas relevant ex-
planations are provided to the users for each of them (via
the chatbot’s User Interface - UI). The questions refer to
basic license terms, including whether modifications to the
software are allowed (question 1), whether the use of a dif-
ferent compatible license instead of the license of the origi-
nal software is allowed for derivative works (question 4 - for
instance, this is allowed under the MIT License but not un-
der GPL licenses), and whether interactions over a network
are covered (question 7 - covering mainly AGPL licenses).
Other requests, e.g. asking for information for specific
licenses, including permissions and restrictions or asking
for more general aspects of OSS licenses, are processed by
the chatbot. In order to answer a user’s request, the chat-
bot performs the next steps after the user types a message.
These are typical steps found in existing chatbot platforms,
such as Rasa or DialogFlow:
10
• Input Pre-Processing. The message of the user is pre-
processed, including text normalizing procedures, such
as converting to small letter cases, removal of punctua-
tion, as well as correction of mistakes in the message.
• Intent Recognition. Natural Language Understanding
(NLU) techniques are used, utilizing machine learning
10
https://cloud.google.com/dialogflow
ENASE 2025 - 20th International Conference on Evaluation of Novel Approaches to Software Engineering
576

Figure 1: General workflow of the Licy license assistant chatbot.
Table 3: Chatbot questions for license recommendations for
advanced users.
1. Do you want the license to allow modifications to your
software?
2. Do you want to allow the licensed software and deriva-
tives to be used for commercial purposes?
3. Do you want to allow your software to be used in propri-
etary (closed-source) applications?
4. Do you want to allow sublicensing of your software, e.g.
addition of a different license?
5. Do you want modifications to your software to be re-
leased under the same license (when distributing the li-
censed software)?
6. Do you want to specify that any modifications made to
your software must be indicated when redistributed?
7. Do you want to allow users that interact with the licensed
software via a network to receive a copy of the source
code?
8. Are you concerned about copyright, patents, or trade-
marks in your work?
9. Do you want the original copyright to be retained?
10. Do you want to give users the right to practice patent
claims of contributors to the code?
11. Do you want to allow the use of trademarks in your
work?
models that have been trained to analyze the text and
classify the user’s intent.
• Entity Extraction. Entities are extracted from the user’s
message. Entities are important details and names men-
tioned in the user’s message, like names of software li-
censes, license category, or other relevant information.
• Dialogue Management. The dialogue manager deter-
mines which response should be provided to handle
user’s input by considering the context of the conver-
sation and the training data of the chatbot.
• Action Selection. An action to be performed next is cho-
sen in this step. Actions can be to return a response to
the user, to ask the knowledge base or to invoke a cus-
tom action. More details on the chatbot’s knowledge
base that drives this step are provided in the next sec-
tion.
• Response Generation. A reply is formed using tem-
plates or responses defined in the training data. The re-
sponse is customized in line with the user’s request and
utilizes any information which was found in the user’s
message.
• Follow-Up Actions. It is determined whether any ad-
ditional actions are needed after performing an action.
This is done by asking the user to provide more details
or revising the user’s query. For instance, the chatbot
may ask the user to verify the license name he/she is re-
ferring to (to ensure it was correctly recognized by Licy
as users may use slightly modified terminology to refer
to a specific license).
3.3 Chatbot’s Training Data and
Knowledge Base
In order to assist the chatbot in performing the steps de-
scribed previously, relevant data have been created. We
have created initially, training data that can be used to
understand user’s intent (Intent Recognition step in sec-
tion 3.2). The data have been created taking into considera-
Licy: A Chatbot Assistant to Better Understand and Select Open Source Software Licenses
577

tion the type of questions that can be asked for each license,
using as basis the license modeling (assuming that users will
focus mainly on these aspects of the license text), and the
variations in the license names that may be employed.
The above are reflected in the knowledge base of Licy,
whose main goal is to cover a wide range of user queries
in the area of license information and license recommen-
dations. Via the knowledge base, the chatbot is able to
understand the intention of user’s inputs. Most chatbots
are designed around intents. The license modeling has
been used in order to create the possible intents, combin-
ing different license names (and variations) with what the
license terms indicate. Some examples of potential in-
tents of the user when asking about licenses are provided
next. Each case corresponds to a specific user’s intent,
with the text in square brackets ([]) corresponding to the
entity’s text (parameter from the user’s text) and the text
in parenthesis (()) corresponding to the entity’s name, e.g.
[MIT](license name). The first example provided concerns
variants of user’s intent to ask for information about a spe-
cific license, while the second variants of asking about a
specific permission a specific license may offer. The com-
plete list of options is available in the software repository
of Licy (Licy, 2024). The scope of what the chatbot can
understand and how it should respond to different user in-
puts is captured in a separate part of the configuration, so
the intents and entities shown below are indicated in this
configuration (domain.yml in Rasa framework):
− i n t e n t : a s k f o r l i c e n s e
e xamp l e s : |
−What do you know a b o u t [ AGPL− 3 . 0 ] ( l i c e n s e n a m e )
−G iv e me some i n f o on [ Vim ] ( l i c e n s e n a m e )
−G iv e me some i n f o on [CC−SA ] ( l i c e n s e n a m e )
−G iv e me some i n f o on [ EU D a t ag r i d ] ( l i c e n s e n a m e )
− I w an t t o know more a b o u t [ U n l i c e n s e ] ( l i c e n s e n a m e )
− I w an t t o know more a b o u t [ EPL − 1 . 0 ] ( l i c e n s e n a m e )
− I w an t t o know more a b o u t [ LPPL − 1 . 3 c ] ( l i c e n s e n a m e )
−What i s [CC−BY− 4 . 0 ] ( l i c e n s e n a m e )
−Have you h e a r d a b o u t [ EUPL − 1 . 1 ] ( l i c e n s e n a m e )
−[ A r t i s t i c − 2 . 0 ] ( l i c e n s e n a m e )
−[BSD−3− C l a u s e ] ( l i c e n s e n a m e )
[ . . . ]
− i n t e n t : a s k f o r l i c e n s e p e r m i s s i o n
e xamp l e s : |
−Does [ CPL − 1 . 0 ] ( l i c e n s e n a m e ) [ a l l o w ] ( c h o i c e )
[ p r i v a t e − u s e ] ( p e r m i s s i o n ) ?
− I s [ s u b l i c e n s i n g ] ( p e r m i s s i o n ) [ a l l o w e d ] ( c h o i c e )
f o r t h e [ MIT ] ( l i c e n s e n a m e ) ?
−Does [ GPL − 2 . 0 ] ( l i c e n s e n a m e ) [ p e r m i t ] ( c h o i c e )
[ m o d i f i c a t i o n s ] ( p e r m i s s i o n ) t o t h e s o u r c e co d e ?
−Does [ APSL − 2 . 0 ] ( l i c e n s e n a m e ) [ l e t ] ( c h o i c e ) t h e
l i c e n s e d m a t e r i a l an d d e r i v a t i v e s t o b e u s e d f o r
[ c o m m e r c i a l p u r p o s e s ] ( p e r m i s s i o n ) ?
−[ Can ] ( c h o i c e ) I [ mo d ify ] ( p e r m i s s i o n ) t h e s o u r c e co d e
o f [ CC−BY−NC] ( l i c e n s e n a m e )
−[ Can ] ( c h o i c e ) I u s e [ c o m m e r c i a l l y ] ( p e r m i s s i o n )
t h e s o f t w a r e w i t h t h e [ E c l i p s e P u b l i c L i c e n s e 1 . 0 ]
( l i c e n s e n a m e ) ?
−Does [ CC−BY−NC] ( l i c e n s e n a m e ) [ f o r b i d ] ( c h o i c e )
[ s u b l i c e n s e ] ( p e r m i s s i o n ) ?
−[EUPL − 1 . 2 ] ( l i c e n s e n a m e ) [ den y ] ( c h o i c e )
[ co m m er c i a l − u s e ] ( p e r m i s s i o n ) ?
[ . . . ]
In addition to intents, flows are used to define exam-
ple conversations that a user might have with the chatbot.
These flows guide the chatbot on how to respond to dif-
ferent user inputs based on the sequence of intents and ac-
tions. Essentially, they serve as training data for the chatbot,
helping it to learn how to manage dialogues by predicting
the next action based on the current state of the conversa-
tion (Dialogue Management and Action Selection steps in
section 3.2). Each flow represents a potential path that a
conversation can take, capturing various user interactions
and the chatbot’s responses. By providing example con-
versations (or stories in the terminology of Rasa), the Rasa
framework trains its dialogue model to understand and pre-
dict the appropriate actions to take during real user inter-
actions. Each step in a flow includes an intent that rep-
resents the user’s message or action. Some example sto-
ries (i.e. flows) used in the chatbot are listed below (as for
intents, the whole list is available in the online repository
of Licy) (Licy, 2024). The action defines the chatbot’s re-
sponse to that intent, which can be a simple reply, a custom
action, or any other predefined response. The names of the
intents indicated correspond to intents already defined in the
knowledge base, as in the previous intent examples:
− s t o r y : Ask f o r s o f t w a r e l i c e n s e d e f i n i t i o n
s t e p s :
− i n t e n t : a s k f o r l i c e n s e d e f i n i t i o n
− a c t i o n : u t t e r l i c e n s e d e f i n i t i o n
− s t o r y : U s er a g r e e s on l i c e n s e
s t e p s :
− i n t e n t : a s k f o r l i c e n s e
− a c t i o n : a c t i o n g e t l i c e n s e
− i n t e n t : a f f i r m
− a c t i o n : a c t i o n g e t l i c e n s e i n f o
− s t o r y : U s er a s k s f o r l i c e n s e p e r m i s s i o n s
s t e p s :
− i n t e n t : a s k f o r l i c e n s e p e r m i s s i o n
− a c t i o n : a c t i o n c h e c k p e r m i s s i o n
3.4 Implementation Details
As aforementioned, the Rasa framework was employed for
implementation purposes, that uses Python, markdown and
YAML. The intents examples are used by the NLU al-
gorithms of Rasa, and are defined in the respective file
(nlu.yml) with the training data that assist Rasa in identify-
ing users’ intents. Flows in the conversation are defined in
the stories.yml file. The machine learning models present in
Rasa are trained using the knowledge base on how to com-
prehend human language. For implementation purposes, all
license terms (for the license modeling of section 3.1) were
encoded in JSON – JavaScript Object Notation – format. In
addition, the Firebase
11
cloud database was used for data
storage purposes, and Flask
12
Python web framework was
used for back-end routing. Standard web technologies were
employed for the front-end implementation, along with a
number of Python libraries used for NLU and Natural Lan-
guage Processing (NLP), e.g. PyPI.
For demonstration purposes, two screenshots from the
prototype of Licy are depicted in Figure 2 and Figure 3,
with the first showing a case of asking information on a
specific license and the second triggering the questionnaire
on license recommendations. From the first example,
the chatbot extracts two main parameters: the license
name (GPL2) and what the user is interested in (info).
The chatbot provides also the possibility to display more
information on the licenses, including relevant links (e.g.
TLDRLegal website).
Chatbot Availability. The chatbot source code and instruc-
tions for its installation are available in a software reposi-
tory online (Licy, 2024). The complete knowledge base of
11
https://firebase.google.com/
12
https://flask.palletsprojects.com/en/3.0.x/
ENASE 2025 - 20th International Conference on Evaluation of Novel Approaches to Software Engineering
578

Figure 2: User asking Licy for a specific license.
Figure 3: User asking Licy to recommend a license.
the chatbot as detailed in section 3.3 is also available in the
repository (Licy, 2024).
4 EVALUATION OF LICY
4.1 Metrics Calculation
As a first evaluation step, we used the basic chatbot met-
rics introduced in a prior work (excluding the ones that re-
quire external tools), in order to get an overview of the gen-
eral design of Licy (Ca
˜
nizares et al., 2022). For this pur-
pose, we have transformed the main configuration files of
Rasa (config.yml, nlu.yml, stories.yml and domain.yml) to
the CONGA (ChatbOt modelliNg lanGuAge) chatbot neu-
tral notation (P
´
erez-Soler et al., 2020).
The results with the existing metrics are listed in Ta-
ble 4. Comparing the values with the metrics of the chatbot
dataset used in (Ca
˜
nizares et al., 2022), it can be seen that
Licy covers a wide range of options: there are 14 differ-
ent entities (with the basic ones concerning license infor-
mation and recommendations), a large number of training
phrases and words per training phrase are used in intents
(only Dining-Out chatbot has a higher number of TPI than
Table 4: Basic metrics calculation for Licy (Ca
˜
nizares et al.,
2022).
Name Description Value
INT # intents 14
ENT # user-defined entities 0
FLOW # conversation entry points 12
PATH # different conversation flow paths 12
Intent metrics
TPI # training phrases per intent 70.07
WPTP # words per training phrase 5.43
PPTP # parameters per training phrase 0.71
CPOP # characters per output phrase 121.60
Flow metrics
FACT # actions per flow 1.33
FPATH # conversation flow paths 1.00
CL conversation length 3
Licy and only googleChallenge a higher number of WPTP
among the 12 chatbot designs compared in (Ca
˜
nizares et al.,
2022)), and the characters per output phrase are also high
(only FAQ-RASA-NLU in (Ca
˜
nizares et al., 2022) has a
higher number). The conversation length (CL) is relatively
small, with the chatbot asking the user usually to confirm
the license he/she is referring to, when asking for specific
license information (as in the examples in section 3.3).
4.2 User Evaluation Design
In order to evaluate the usefulness and the overall experi-
ence of Licy, a small scale user evaluation was performed.
Questions introduced focused on assessing the visual appeal
of the user interface, the user-friendliness of the chatbot,
and the overall ease of use. The questionnaire consists of
three main parts: 1) the first part asks users about their ex-
isting knowledge on OSS and OSS licenses, 2) the second
part includes usability questions on the chatbot (e.g. how
easy it was to use it), 3) the final part asks whether the chat-
bot was useful and their opinion on whether the informa-
tion provided is accurate. Users were also asked to indicate
any additional comments they might have, while access to a
chatbot deployment was provided so that users could inter-
act with the chatbot before answering the questionnaire.
In order to recruit participants, emails were sent to
Computer Science students and researchers within Univer-
sity of Cyprus. Users were informed that no personal data
would collected, and that any data used for research pur-
poses would not lead to the participant in any way. In order
to answer the questionnaire, participants gave their consent.
4.3 User Evaluation Results
18 individuals participated in the study, with all of them
being Computer Science students or researchers. Regard-
ing prior knowledge of OSS and OSS licenses (first part of
questionnaire), most users have intermediate knowledge of
OSS and beginner knowledge of OSS licenses (Figures 4
and 5). Concerning usability aspects, participants found
Licy’s interface user friendly (61.1% found it very user
friendly and 33.3% user friendly). They also found buttons
and icons clear and easy to understand (94.4% of partici-
pants), with only one participant (5.6%) commenting that
“the close button for the about us screen was not clear.”
Licy: A Chatbot Assistant to Better Understand and Select Open Source Software Licenses
579

Figure 4: Participants knowledge on OSS.
Figure 5: Participants knowledge on OSS licenses.
The about us screen is part of the additional information
provided in the chatbot, as in the case of text with URLs
pointing to additional information on the OSS licenses sup-
ported. All users also indicated that it was easy or very
easy to use the chatbot and find the information they were
looking for using Licy. Most users (88.9%) found that the
chatbot understood their questions clearly and two users in-
dicated that this happened only sometimes (11.1%).
On the use of the chatbot (third part of the question-
naire), not all users found it helpful and informative: 33.4%
did not (Figure 6). Nevertheless, most said they thought the
chatbot recommended the appropriate license(s) for them
(82.4% replied positively). Even though the number of par-
ticipants is small, we run statistical tests in order to examine
tendencies in the data that larger datasets should verify. We
run a Kruskal-Wallis H rank-based non-parametric test, us-
ing 1) whether the users found the chatbot helpful and infor-
mative and 2) whether they found the chatbot recommended
the appropriate license as the dependent variables, and the
participants prior knowledge on a) OSS and b) OSS licenses
as independent variables (Vargha and Delaney, 1998). No
statistically significant results were found, indicating that
the prior knowledge of users on OSS and OSS licenses does
not affect their experience with Licy. The same conclusion
was drawn using as dependent variable 1) whether it was
easy to use the chatbot, and 2) whether the user can easily
find the information he/she is looking for.
In the participants’ additional comments, where they
were also asked to indicated additional features they would
find useful, some useful additional features emerged, in-
cluding the following: “Provide licenses that fit on a more
general idea. Like I am creating an application for logistics
company what licenses should I check/need.”, “Information
such as in which cases it is useful to use a specific license
or real-world problems that require this license would be
helpful.”, “Be capable to understand natural language bet-
ter.”, “I would find it useful if the chatbot responded to the
question “What can I ask you and how”.” The last sugges-
Figure 6: Participants’ opinion on whether chatbot is help-
ful and informative.
tion is already available in Licy, but the user might not have
used a relevant phrase the chatbot can understand to answer
this specific question (for training purposes, 18 alternatives
were provided to Licy).
5 DISCUSSION
For the chatbot implementation, a large number of licenses
was taken into consideration (58 licenses including content
licenses). Thus, the chatbot has a good coverage on
licenses users might be interested in, and can guide
license recommendations among all supported licenses.
Concerning the users comments on additional features
and better understanding of the user’s intent, indeed the
chatbot focuses on specific aspects – license details and
license recommendations based on the employed license
model – so it is expected that the users would prefer to
have more options as the user evaluation showed. For
novice users, it is more preferable to provide examples
of licenses for specific cases, without potentially going
into the details of the license terms. Nevertheless, the
current implementation of Licy can be useful primarily for
educational purposes, and can act as a guide for future
and junior software engineers. Its integration in Computer
Science courses or educational material for junior software
engineers could be considered as a useful implication of the
current work, and we argue that it may be more appropriate
in this setting, considering that at the current stage we have
not performed an evaluation with software engineers from
the industry (Ait Baha et al., 2024). We also argue that the
chatbot can be used as a starting point for creating relevant
conversational assistants on OSS licensing.
Limitations. The chatbot’s knowledge base is wide but it
does not cover all Open Source Software licenses avail-
able today (as aforementioned, SPDX lists 634 OSS li-
censes). Moreover, the chatbot’s license recommendations
are generic and rely on the text of licenses and the specific
terms as captured in the modeling of choosealicense, so
they may not address all legal concerns a user may have.
As such, the chatbot cannot be used to provide legal ad-
vice but offers guidance and recommendations that can be
later verified via the help of a legal expert. User evalua-
tion was limited to the users sampled that included students
and researchers and does not include feedback and experi-
ences from a wider population. Participants were mainly
young individuals and thus, users with limited experience.
The participation of senior researchers or developers from
ENASE 2025 - 20th International Conference on Evaluation of Novel Approaches to Software Engineering
580

the industry may have led to different conclusions.
6 CONCLUSIONS
In this work, we have introduced Licy, a chatbot assistant
for OSS licensing. We have described the design and imple-
mentation process of Licy that currently covers 58 OSS li-
censes. The chatbot has been evaluated with chatbot design
metrics and via a small user population, showing promising
results for the future. As future work, we intend to utilize
LLMs to handle cases, where the chatbot is not able to re-
ply appropriately or does not understand the text provided
by the user. They have not been integrated in the current
state, as we wanted to focus initially on providing dedicated
training data specific to OSS licenses and their modeling.
We also aim to evaluate the chatbot with developers from
the industry and expand the cases addressed to include ad-
ditional licenses and multi-licensing schemes.
REFERENCES
Ait Baha, T., El Hajji, M., Es-Saady, Y., and Fadili, H.
(2024). The impact of educational chatbot on student
learning experience. Education and Information Tech-
nologies, 29(8):10153–10176.
Ca
˜
nizares, P. C., P
´
erez-Soler, S., Guerra, E., and De Lara,
J. (2022). Automating the measurement of het-
erogeneous chatbot designs. In Proceedings of the
37th ACM/SIGAPP Symposium on Applied Comput-
ing, pages 1491–1498.
Clarizia, F., Colace, F., Lombardi, M., Pascale, F., and
Santaniello, D. (2018). Chatbot: An education sup-
port system for student. In Cyberspace Safety and
Security: 10th International Symposium, CSS 2018,
Amalfi, Italy, October 29–31, 2018, Proceedings 10,
pages 291–302. Springer.
Dam, S. K., Hong, C. S., Qiao, Y., and Zhang, C. (2024).
A complete survey on llm-based ai chatbots. arXiv
preprint arXiv:2406.16937.
Franciscatto, M. H., Fabro, M. D. D., Trois, C., Cabot, J.,
and Gonc¸alves, L. A. O. (2022). Querying multidi-
mensional big data through a chatbot system. In Pro-
ceedings of the 37th ACM/SIGAPP Symposium on Ap-
plied Computing, pages 381–384.
Gobeille, R. (2008). The fossology project. In Proceed-
ings of the 2008 international working conference on
Mining software repositories, pages 47–50.
Kapitsaki, G., Paphitou, A., and Achilleos, A. (2022).
Towards open source software licenses compatibility
check. In Proceedings of the 26th Pan-Hellenic Con-
ference on Informatics, pages 96–101.
Kapitsaki, G. M. and Charalambous, G. (2016). Find your
open source license now! In 2016 23rd Asia-Pacific
Software Engineering Conference (APSEC), pages 1–
8. IEEE.
Kapitsaki, G. M. and Charalambous, G. (2019). Model-
ing and recommending open source licenses with find-
osslicense. IEEE Transactions on Software Engineer-
ing, 47(5):919–935.
Kapitsaki, G. M. and Kramer, F. (2014). Open source li-
cense violation check for spdx files. In Software Reuse
for Dynamic Systems in the Cloud and Beyond: 14th
International Conference on Software Reuse, ICSR
2015, Miami, FL, USA, January 4-6, 2015. Proceed-
ings 14, pages 90–105. Springer.
Laurent, A. M. S. (2004). Understanding open source and
free software licensing: guide to navigating licensing
issues in existing & new software. ” O’Reilly Media,
Inc.”.
Lebeuf, C., Storey, M.-A., and Zagalsky, A. (2017). Soft-
ware bots. IEEE Software, 35(1):18–23.
Licy (2024). https://github.com/CS-UCY-SEIT-lab/Licy
License Chatbot V2.
Lin, K.-J., Lin, Y.-H., and Ko, T.-M. (2009). Examin-
ing open source software licenses through the cre-
ative commons licensing model. In Software Appli-
cations: Concepts, Methodologies, Tools, and Appli-
cations, pages 2978–2990. IGI Global.
Lokman, A. S. and Ameedeen, M. A. (2019). Modern chat-
bot systems: A technical review. In Proceedings of the
Future Technologies Conference (FTC) 2018: Volume
2, pages 1012–1023. Springer.
Moore, R. J. and Arar, R. (2018). Conversational ux design:
an introduction. Studies in conversational UX design,
pages 1–16.
P
´
erez-Soler, S., Guerra, E., and De Lara, J. (2020). Model-
driven chatbot development. In International Con-
ference on Conceptual Modeling, pages 207–222.
Springer.
Reddy, H. R. (2009). Jacobsen v. katzer: the federal circuit
weighs in on the enforceability of free and open source
software licenses. Berkeley Tech. LJ, 24:299.
Sassi, S. B. (2024). Oslife-disc: Open source licenses
discovering, selecting and comparing. SoftwareX,
27:101761.
Shinde, N. V., Akhade, A., Bagad, P., Bhavsar, H., Wagh,
S., and Kamble, A. (2021). Healthcare chatbot system
using artificial intelligence. In 2021 5th International
Conference on Trends in Electronics and Informatics
(ICOEI), pages 1–8. IEEE.
Stewart, K., Odence, P., and Rockett, E. (2010). Software
package data exchange (spdx) specification. IFOSS L.
Rev., 2:191.
Vargha, A. and Delaney, H. D. (1998). The kruskal-wallis
test and stochastic homogeneity. Journal of Educa-
tional and behavioral Statistics, 23(2):170–192.
Xu, A., Liu, Z., Guo, Y., Sinha, V., and Akkiraju, R. (2017).
A new chatbot for customer service on social media.
In Proceedings of the 2017 CHI conference on human
factors in computing systems, pages 3506–3510.
Xu, W., Wu, X., He, R., and Zhou, M. (2023). Licenserec:
Knowledge based open source license recommenda-
tion for oss projects. In 2023 IEEE/ACM 45th Inter-
national Conference on Software Engineering: Com-
panion Proceedings (ICSE-Companion), pages 180–
183. IEEE.
Licy: A Chatbot Assistant to Better Understand and Select Open Source Software Licenses
581