The rise of the (modelling) bots:Towards assisted modelling via Social Networks

Sara Perez-Soler, Esther Guerra, Juan de Lara, Francisco Jurado

MISO - Modelling & Software Engineering Research Group (miso.es)Universidad Autonoma de Madrid (Spain)

Esther Guerra Assisted modelling via Social Networks ASE 2017 1 / 9

Motivation

Modelling applications are heavyweight (desktop, diagramming)

Social networks, like Twitter and Telegram:

increasingly used for work and leisureagile and lightweight means for coordination and information sharing

Goal: exploit social networks for collaborative modellingsocial networks as front-end for modellingdomain requirements as short messages in natural languagea bot interprets the messages and derives a domain model

Esther Guerra Assisted modelling via Social Networks ASE 2017 2 / 9

ApproachInteraction via messages

Houses have windowsNL description

(ROOT(S

(NP (NNS houses))(VP (VBP have)

(NP (NNS windows)))))

parse tree

users existingmodel

parse

extractactions

add

delete

connect

actions

updatemodel

traceabilitymodel

producefeedbackfeedback

2

4

…

message

kind?

Mike! That is wrong!comment1a

selectNL rule(s)

NL processing rules

3

add class HouseNL command1c

WordNetmodelupdate

projects process1bmanagement

command

feedbacksocial network

social network

Users interact by sending messages to social network of choice.

Discussion and coordination via regular messages

Project management commands (e.g., projects)

Domain requirements in natural language:

descriptions: “houses have windows”

flexible commands: “add class house”, “create class house”

Esther Guerra Assisted modelling via Social Networks ASE 2017 3 / 9

ApproachProcessing of messages in natural language

Houses have windowsNL description

(ROOT(S

(NP (NNS houses))(VP (VBP have)

(NP (NNS windows)))))

parse tree

users existingmodel

parse

extractactions

add

delete

connect

actions

updatemodel

traceabilitymodel

producefeedbackfeedback

2

4

…

message

kind?

Mike! That is wrong!comment1a

selectNL rule(s)

NL processing rules

3

add class HouseNL command1c

WordNetmodelupdate

projects process1bmanagement

command

feedbacksocial network

social network

Bot parses the message (Stanford parser)

Rulesa to interpret parse tree and trigger model update actions

A picture of the updated model is sent to users

a “Extracting domain models from NL requirements: approach and industrial evaluation”, Arora et al., MODELS 2016.

Esther Guerra Assisted modelling via Social Networks ASE 2017 3 / 9

ApproachFeedback and traceability

Houses have windowsNL description

(ROOT(S

(NP (NNS houses))(VP (VBP have)

(NP (NNS windows)))))

parse tree

users existingmodel

parse

extractactions

add

delete

connect

actions

updatemodel

traceabilitymodel

producefeedbackfeedback

2

4

…

message

kind?

Mike! That is wrong!comment1a

selectNL rule(s)

NL processing rules

3

add class HouseNL command1c

WordNetmodelupdate

projects process1bmanagement

command

feedbacksocial network

social network

Model validation

Exporter to Ecore/EMF

Trace model

Esther Guerra Assisted modelling via Social Networks ASE 2017 3 / 9

Example

a goods transportcompany handlesdeliveries

Esther Guerra Assisted modelling via Social Networks ASE 2017 4 / 9

Example

a goods transportcompany handlesdeliveries

a delivery has anumeric identifier

⇒

Esther Guerra Assisted modelling via Social Networks ASE 2017 4 / 9

Example

a goods transportcompany handlesdeliveries

a delivery has anumeric identifier

⇒

a delivery is made ofpackages. Packets can bebulky, heavy or fragile

⇒

Esther Guerra Assisted modelling via Social Networks ASE 2017 4 / 9

Key pointsBenefits

Social networks are ubiquitous (low learning curve)

Use in mobility, no need to install new applications

Lightweight front-end

Interaction via short messages can be easier/faster

Requirements in natural language (suitable for non-modelling experts)

A bot interprets the messages and derives a model

Seamless integration of modelling and discussion mechanisms

Message history provides trace information

Esther Guerra Assisted modelling via Social Networks ASE 2017 5 / 9

Key pointsScenarios

Quick prototyping of models when and where needed

Sw projects: foster collaboration of engineers and domain experts

Education: collaborative resolution of exercises

Crowdsourcing of modelling decisions

Esther Guerra Assisted modelling via Social Networks ASE 2017 5 / 9

Tool support

SOCIO is a bot for assisted modelling over social networks

It works over Telegram and Twitter (@modellingBot)

It uses the Stanford parser (parsing) and Wordnet (synonyms)

Video and URL: https://saraperezsoler.github.io/ModellingBot

Telegram Twitter

Esther Guerra Assisted modelling via Social Networks ASE 2017 6 / 9

Evaluation

Participants: 10 post-graduate or last-year degree students of CS

Configuration: 4 Telegram groups (2-3 people/group)

Task: building model for e-commerce, answering questionnaire

Results:

good usability (74%)natural language is suitable to build models (75%)social networks are useful for collaboration (76%)easy-to-use, quicker than other modelling tools

Esther Guerra Assisted modelling via Social Networks ASE 2017 7 / 9

Summary and next steps

Novel approach to collaborative modelling via social networks

Tooling and promising initial evaluation

What’s next?

improve natural processing, extend command set of tooldefine collaboration protocols (e.g., roles, voting)other bots (e.g., quality assurance bots, gamification bots)other social networks (e.g., Slack) and communication mechanisms(e.g., speech recognition using Skype bots)

Esther Guerra Assisted modelling via Social Networks ASE 2017 8 / 9

The rise of the (modelling) bots:Towards assisted modelling via Social Networks

Sara Perez-Soler, Esther Guerra, Juan de Lara, Francisco Jurado

MISO - Modelling & Software Engineering Research Group (miso.es)Universidad Autonoma de Madrid (Spain)

Questions?

Esther Guerra Assisted modelling via Social Networks ASE 2017 9 / 9