36th International Symposium on Automation and Robotics in Construction (ISARC 2019)

A Chatbot System for Construction Daily Report

Information Management
Jehyun Choa and Ghang Leea
a
Department of Architecture and Architectural Engineering, University of Yonsei, South Korea
E-mail: [email protected], [email protected]

Abstract – prevents them from using it. Nevertheless, the use of

Consistently updating, analyzing, and managing everyday instant messaging (IM) applications at
construction-related information is one of the key construction sites has been on the rise as a channel of
success factors in project management. Quite a few information exchange. There are many advantages to
construction projects have recently started to utilize using popular IM applications, such as interoperability
instant messaging (IM) applications such as Slack, with external systems, searching for information, and
WhatsApp, and WeChat as a communication channel exchanging files.
among project participants to share daily In addition, chatbot technologies can be deployed. A
construction information due to easy accessibility. chatbot is a service that provides a communication
However, general contractors are still required to interface with users based on rules or artificial
manually extract and integrate the data from instant intelligence (AI). An AI-based chatbot answers based on
messages to compose daily reports. This is because the the understanding the intentions of texts from users using
data inputted by subcontractors through IM natural language processing (NLP), while rule-based
applications are usually in an unstructured form and chatbot can answer only when detect predefined
the IM application is not normally interoperable with expressions[1]. When using chatbot technologies that
the systems database especially developed for understand intent and context from a user’s utterances,
construction management. To solve this problem, this and provide the necessary information through an
study proposes a chatbot-assisted construction daily interactive interface, there is the potential to support the
report data management system. The chatbot in the user’s efforts to collect and unify information. Therefore,
proposed system collects and processes the required this study presents a chatbot system model that collects
information through conversations with construction daily report data through conversations with
subcontractors, and automatically generates and users, such as general contractors and subcontractors, and
shares a daily report for general contractors. A automatically generates daily report.
prototype system has been designed and implemented This paper is organized as follows. First, the
to prove the concept. requirements of the chatbot system are defined through
the literature review. Then, the system architecture and
Keywords – dialog model for the two main functions of data
Construction daily report; Chatbot; Instant extraction and report generation are explained. Finally,
messaging application the functional feasibility was validated through the
development of prototypes.

1 Introduction
2 Literature Review
Consistently collecting and analyzing construction
management related information, such as the schedule
and cost, is very important for successful project results 2.1 Limitations of Previous Studies on
due to the inherent risks and uncertainty of the Construction Daily Report Systems
construction industry. Therefore, a systematic Many studies have suggested a system model for
communication plan or process between the general troubleshooting time-consuming and inefficient
contractor and subcontractors needs to be defined for processes performed each day for collecting data and
efficient data collection. creating construction daily reports.
However, it can be problematic if all participants do Russell et al. emphasized the importance of readily
not utilize a unified communication channel or if a low and reliably collecting data while presenting an early
level of technical understanding about the system model of the computing system [2]. However, even with
 36th International Symposium on Automation and Robotics in Construction (ISARC 2019)

the advantages of storing and utilizing information, the relevant documents for users.
proposed early system model was not able to relieve the
burden on the business because the manager still had to
input all the information at construction field. In a case 3 Requirement and Coverage Definition
study lead by Yan-chyuan et al., a system model in which of the System
a daily report module extends its functionality by
incorporating the estimation, pricing, and accounting 3.1 Requirement of the System
system to enable cost management [3]. However, it was
also not clearly defined for the main body to enter the The system requirements defined through the
necessary report data. literature review are as follows:
Subsequently proposed system models were designed
to allow subcontractors to enter construction daily report 1. The chatbot system should be able to extract the
data. The system model designed by Chin et al. would information related to the daily report from the
reduce a manager’s tasks and enable plan-completion messages or through the conversation with the
management by entering detailed level of work items by chatbot and automatically store it in a linked
subcontractors [4]. More recently, a number of BIM- database.
based cloud computing software programs, such as 2. The chatbot system should be able to generate
BIM360 and Procore, allow for effective construction reports automatically based on information from the
project management by reflecting the system model database.
explained above.
However, while this software provides strong With respect to the first requirement, it is necessary
management functionality, it has a major issue in that the to define which types of information should be handled
systems require users to have a high level of technical by this system. In this research, information types are
understanding. Several factors, such as lack of education defined based on prior research by Song [6]. Song has
programs and low levels of subcontractor technical analyzed conversations that have taken place using IM
expertise, have been cited as the main cause of the applications on seven actual construction projects.
problem with regards to using a unified platform [5]. Through the analysis, “date,” “name of company,” “types
and number of labor,” “types and number of equipment,”
and “work items” have been defined as primarily shared
2.2 Possibilities and Limitations in Current
information types through IM applications. Figure 1 is an
Use of IM Applications for Daily Report example message for daily reporting, including the five
Information Collection information types used in construction projects.
Recently, participants of quite a few construction With respect to the second requirement,
projects have communicated using IM applications automatically created documents are defined as including
originally developed for nonbusiness purposes [6]. “project name,” “date,” “total number of units per labor
IM applications can provide better management type,” “total number of units per equipment type,” and
capabilities when coupled with external management “work item list” as shown in Figure 2.
systems, such as a BIM-based construction management In order to calculate the total number of laborers and
system explained above. However, due to the lack of equipment used, the sum of the numbers entered by
connectivity, managers are currently required to spend
additional time reformatting, integrating, and re-entering
the same information.
This study proposes a “chatbot” as a solution to such
redundant and manual data-input problems. In the
construction domain, NLP has been mainly used for
analysis of documentations for checking code
compliance [7], retrieving similar project cases [8], and
managing safety factors [9]. However, there are no
studies utilizing NLP for construction daily report related
data management.
In order to automatically manage information related Figure 1. Mainly shared information types in an
to the construction daily report through communications example user message for daily reporting
with the chatbot, the specific model of the chatbot system
needs to be designed to understand the messages from the different subcontractors must be determined.
users, to automatically extract the necessary information
or store it in the database, and to automatically generate
 36th International Symposium on Automation and Robotics in Construction (ISARC 2019)

3.2 Coverage of the Defined Information The architecture of the chatbot system to meet the
Types in the System defined requirements will be explained in the next section.
In the case of labor and equipment, it has been
determined that it is generally entered in a structure that 4 Design of the Chatbot System
follows the order of “type,” “quantity,” and “unit name,”
such as “Mason 6 people” in Figure 1. Therefore, in this The architecture of the system to meet the above
requirements is designed as shown in Figure 3. This
study, a regular expression (regexp) approach is applied
system consists of a web application, a chatbot module,
to detect the types “number of labor” and “number of
and a back-end module that connects to IM applications.
equipment.”
The web application receives the text input from the
user through the IM application and transfers it to the
chatbot module. The chatbot module then analyzes the
messages from the users and generates a responding
sentence or call using several functions of the web
application and utilizing the back-end module.
The chatbot is an AI module based on Natural

Figure 2. defined report format and included

information types
In defining the regular expression for the “number of Figure 3. The architecture of the proposed system
labor” and “number of equipment,” because the detailed
Language Processing (NLP) and it plays a role in
items of both types may vary depending on the
understanding sentences that are received from users and
breakdown structure applied, the items in Table 1 are
then producing appropriate responses. The chatbot
temporarily defined as labor and equipment items that the
module is composed of “intent,” “entity,” and “dialog.”
system deals with.
Intent represents the purpose of the input by a user. All
types of requests users made by users are defined as
Table 1. Temporary breakdown structure of labor and intent.. Entity means a term or object related to intent.
equipment covered by the chatbot system Entity is used to provide a specific context for intent.
Equipment Dialog is a branching conversation flow that defines
Labor types responses to the defined intents and entities.
types
Carpenter Electrician Crane The web application consists of a “data extractor” and
Supervisor Mason Loader “report generator.” Both modules are activated by
function calls derived from the analysis results of chatbot
Ironworker Plumber Backhoe modules.
Glazier Dozer The back-end system includes a database for storing
Operator Welder extracted daily construction information obtained
Pipelayer Pumpcar through message analysis and a file system for managing
Truck daily construction documents that are generated
Compressor
driver automatically upon user request.
Dump In the following section, the design of the dialog
Plasterer
Truck model, intents, and entities for information input and
Individual input about work items for the day is report generation in the chatbot module are explained.
designed to be made through a dialog with the chatbot
because it is determined that a phrase about work items
for the day follows a declaration expression such as
“work item” in Figure 1 and is not entered in a structured
form.
36th International Symposium on Automation and Robotics in Construction (ISARC 2019)

Figure 4. Dialog design for information input

 36th International Symposium on Automation and Robotics in Construction (ISARC 2019)

The dialog model includes example messages messages as designed in Table 3. The remaining four
between the user and the chatbot. Messages from the information types were defined as “entities” for this
chatbot are generated by detecting one of the predefined dialog model, except for the non-structural form of work
“intents” and “entities” based on the context of the dialog. item mentioned previously.
The intent type “date” borrowed the most common
4.1 Chatbot Module Design for Information form of structure, and the intent type “name of company”
Extraction is defined as matching the text within a list of company

First, the dialog model for information extraction has

been designed as illustrated in Figure 4.
The conversation begins with the chatbot detecting a
message, representing the intention of the user to input
information. Subsequent conversation flows are designed
for chatbot to ask users for information in the order of
labor, equipment, and work item. Finally, it is designed
for the user to receive a final confirmation message
generated by summarizing the information.

Table 2. Defined intents and examples of user

utterances
Intent Example expressions
“I want to upload information for
Information Figure 5. Designed regular expression approach
input the daily report” for entity detection
“Information input for daily report”
names.
Positive
“Yes, please,” “Great” Intent types “number of labor” and “number of
answer
equipment” takes the identical approach with regards to
the pattern. Figure 5 describes the structure design of the
Negative answer “No, it isn’t,” “No” regular expressions for recognizing the example text for
the number of equipment. The pattern structure consists
The types of intent to cover the dialog model defined of “equipment type,” “number of quantity,” and “name
above are given in Table 2. All the example expressions of unit.” The regular expression structure also includes
from users are necessary for the NLP training of the AI consideration for “singular and plural form,” “existence
chatbot. The intent “information input” is designed to of space,” “upper case and lower case,” and “multiline”
start a conversation process about information extraction for the text from users.
with the chatbot. The “positive answer” and “negative Although Figure 4 explains the longest possible flow
answer” are designed for the confirmation step in the last of conversation, based on the defined types of entities, it
part of dialog. is also possible to have a scenario in which the chatbot
module extracts all of the information for each of the four
Table 3. Defined entities and value types types from a single message sent by the user.
Entity Value type Structure description
4.2 Chatbot Module Design for Report
Combination of year, Generation
Date Patterns
month, and day
The dialog model for daily report generation has been
Name of Synonym from designed as illustrated in Figure 6. The conversation
Synonyms
company predefined list begins with the chatbot detecting a message representing
Number of Combination of type, the intention of the user to request a document.
Patterns Subsequent conversation flows are designed for the
labor number of quantity,
name of unit chatbot to ask for the user’s e-mail address to send the
Number of generated document. The conversation is finished when
Patterns (Figure 5)
equipment the user receives an e-mail with the generated daily report.
The types of entity are defined in order to detect the This dialog model is designed for short conversations;
types of information and extract them from the user only one type of intent “requiring document” is designed.
 36th International Symposium on Automation and Robotics in Construction (ISARC 2019)

Figure 6. Dialog design for daily report generation

Several example expressions, including the message to determine how well it satisfies the requirements
“Please send me today’s construction report,” are identified in the requirements analysis phase of system
required for AI training as shown in Figure 6. development.
A single entity type is defined to obtain an e-mail First, the collaborative platform Slack, a well-
address from a user in regular expression pattern. established IM application, has been utilized for
It is also possible to skip this dialog in actual system connecting users and the chatbot system.
implementation because once the system saves the email The chatbot module has been implemented with
address, or has alternative information such as a phone Watson Assistant, which is an AI as a Service (AIaaS)
number, the chatbot can automatically generate the provided by IBM for designing and implementing
documents and send them to the user on a regular basis chatbots.
at a defined time. The web application has been implemented as a
Node.js application. Node.js is an open source
development platform for executing JavaScript code
5 Prototype Implementation server-side. In the development of the Node.js web
application, several packages shared through NPM
5.1 Configuration of the Prototype System (Node Package Manager) are included as dependencies
A prototype of the proposed system as depicted in of the system as explained in Table 4.
Figure 7 was developed to check the functional feasibility
Table 4. Utilized packages for Node.js web application
in the prototype
Name Role
Integrating IM application and
botkit
chatbot service
Providing web application
Express
framework
Node-
Handling database
MySQL
Generating and updating excel
XLSL
file
Nodemailer Sending e-mail
Figure 7. Configuration of the system
prototype The “botkit” package has been utilized to integrate
 36th International Symposium on Automation and Robotics in Construction (ISARC 2019)

Figure 8. Implementation of the dialog models in the prototype system

the IM application and the chatbot service, which are First, as depicted in area A of Figure 8, it is confirmed
Slack and Watson Assistant in the prototype. The that, except for the “date” information, information
“Express” package has been utilized for operating the related to the “name of company,” “number of labor,”
server. The “Node-MySQL” package has been utilized “number of equipment,” and “work items” was properly
for handling the database. The “XLSX” package has been extracted from user messages, saved in the form of
utilized to generate and update the Excel file for the daily context variables during the process of conversation, and
report. The “Nodemailer” package has been utilized to finally utilized for generating confirmation message.
send emails to users. Following this, as described in area B of Figure 8,
The database system has been implemented with user-confirmed data was parsed and stored by the web
MySQL, which is a popular open source database. application as multiple rows in the database. In the case
of “number of labor” and “number of equipment,”
5.2 Implementation of the Designed Dialog quantity information was extracted and stored separately
Model in a Prototype System from the structured text such as “8” in the example
“Carpenter 8 people,”.
Data collection and automated report generation were Finally, as shown in area C of Figure 8, a document
tested using the prototype system based on predefined was automatically generated by the web application by
dialog models. entering the information from the database into the
 36th International Symposium on Automation and Robotics in Construction (ISARC 2019)

report’s Excel form, which is predefined in advance for subcontractors in terms of ease of use and for general
when requested by the user. contractors with regards to reducing the effort to collect
and document information. In addition, studies required
5.3 Discussion for structural recognition of work items and functional
range extensions for other management purposes, such as
Through the experiments, the chatbot system has schedule and cost management, are further expected.
been identified as a functional and feasible system model
that is expected to provide the following benefits:
· It can serve as a basis for enhancing management
Acknowledgements
functions because IM application data is no longer This work was supported by Institute for Information
volatile; it is now part of the system environment. & communications Technology Promotion(IITP) grant
· This system model is beneficial to both user groups. funded by the Korea government(MSIT) (No.2017-0-
The subcontractors do not require any special 00701(2nd year), Development of the Smart Field
courses in the system input process, and the burden Platform for Planning and Managing Building
of data collection and documentation is reduced for Construction), grant (no. 19AUDP-B127891-03) from
the general contractors. the Architecture and Urban Development Research
Program funded by the Ministry of Land, Infrastructure,
Nevertheless, the proposed system model has some
and Transport of the Korean Government and partially
problems to overcome at the current level.
supported by the Graduate School of Yonsei University
· In the case of labor and equipment, the structure of Research Scholarship Grants in 2017.
text was borrowed from a previous study. However,
for the “work items,” because each subcontractor
has a different work breakdown structure (WBS) References
and the text of the message is mixed with [1] Shridhar K. Rule based bots vs AI bots. Online:
information about the work location in an https://medium.com/botsupply/rule-based-bots-
unstructured form, it needs to incorporate a vs-ai-bots-b60cdb786ffa, Acessed: 2017,
cognitive approach based on standardized WBS and 03/22/2019.
LBS (Location Breakdown Structure). [2] Russell A. D., Computerized daily site reporting,
· Due to the nature of the interactive interface, Journal of Construction Engineering, 119
compared to the visual user interface, it is necessary (2):385-402, 1993.
to extend the dialog model in terms of convenience [3] Shiau Y. and Wang W., Daily report module for
to meet various user scenarios. For example, it may construction management information system,
be possible to modify only some of the total Proc. 20th Int. Symp. on Automation and
information entered, or to re-check the information Robotics in Construction, 2003.
entered. [4] Chin S., Kim K., and Kim Y.-S. J. J. o. c. i. c. e.,
Generate-select-check based daily reporting
system, 19 (4):412-425, 2005.
6 Conclusion [5] Construction M. H., "The Business Value of
This study presented a chatbot system model that BIM for Construction in Major Global
automatically extracts information related to the Markets." McGraw Hill Construction, Bedford,
construction work daily report exchanged in IM MA.," McGraw Hill Construction2014.
applications and automatically generates the construction [6] Song T. S., "Extracting daily report information
daily report. The proposed system model includes a from text messages between the field
chatbot module, web application, and database. Several managers," Master, Architecture &
types of “intents” and “entities” are defined as needed for Architectural Engineering, Yonsei University,
the conversation between users and the chatbot, along 2018.
with dialog models. Detected information through the [7] Zhang J. and El-Gohary N. M., Integrating
conversation between the user and chatbot could be semantic NLP and logic reasoning into a unified
utilized by the web application for data extraction and system for fully-automated code checking,
document generation using a database. Automation in Construction, 73:45-57,
The prototype was developed and validated to 2017/01/01/ 2017.
determine whether the chatbot system can perform [8] Zou Y., Kiviniemi A., and Jones S. W.,
functions related to the requirements of data extraction Retrieving similar cases for construction project
and document generation. risk management using Natural Language
The proposed system is expected to provide benefits
 36th International Symposium on Automation and Robotics in Construction (ISARC 2019)

Processing techniques, Automation in

Construction, 80:66-76, 2017/08/01/ 2017.
[9] Tixier A. J. P., Hallowell M. R., Rajagopalan B.,
and Bowman D., Automated content analysis
for construction safety: A natural language
processing system to extract precursors and
outcomes from unstructured injury reports,
Automation in Construction, 62:45-56,
2016/02/01/ 2016.