Tool and Evaluation Method for Idea Creation Support

Ryo Takeshima and Katashi Nagao

Department of Media Science, Graduate School of Information Science, Nagoya University, Nagoya, Japan

Keywords: Idea Creation, Idea Evaluation, Digital Poster, Meeting Support, Collaboration, Machine Learning.

Abstract: We have developed a new idea creation support tool in which (1) each idea is represented by a tree structure,

(2) the idea is automatically evaluated on the basis of the tree structure so that the relative advantages among

several alternative ideas is found, (3) the ideas are presented in a poster format, and (4) the ideas are shared

by multiple users so that the ideas can be quoted and expanded upon by individual users. In this work, we

explain the mechanisms of this tool, including the evaluation and poster conversion of ideas and collaborative

idea creation, and briefly discuss our plan for the future.

1 INTRODUCTION

Although it is well-known that idea creation and

sharing are beneficial, and a variety of tools for this

purpose have been developed, it is still very difficult

to automatically evaluate the quality of an idea.

Previous idea creation support tools have contributed

to the creation, organization, and visualization of

ideas, but they have not provided a function for

evaluating the created ideas. We have therefore

developed a method to relatively evaluate ideas by

assessing the appropriateness of the data representing

the ideas.

First, our developed tool helps users express the

idea in a tree structure and then creates a presentation

in a poster format on the basis of this structure. The

tool accesses the database of the idea, called the “idea

pool”, to search and share ideas in the tree structures.

The tree structure consists of a title to express the

idea, its description, relevant keywords and their

descriptions, images, Web pages, and videos as nodes

and is defined by the hierarchical relationship

between the nodes. Similar methods to express ideas

by tree or graph structures have been proposed, a

typical example of which is the Mind Map (Buzan,

T., 1990) (Beel and Langer, 2011).

To clearly visualize the idea expressed in a tree

structure, we developed a method of displaying the

idea by a poster representation. This method arranges

the contents of the nodes on the screen on the basis of

the tree structure of the ideas. The created poster is

interactive: any node (a block in the poster) can be

displayed in more detail when the user selects and

requests. This refinement does not merely enlarge the

block—it also compensates for a convenient portion

omitted by the area constraint and customizes the

content to fit the larger area. We call this “semantic

zooming.”

Using poster representations enables an easier

presentation of ideas, and hopefully additional ideas

will be generated by the presentation and subsequent

discussion. The poster can be shared by a group of

users, along with its underlying tree structure. By

reusing the shared poster, additional ideas of the

community members can be integrated with the

original idea.

Evaluating the ideas expressed in a tree structure

makes it possible to prioritize the ideas. When

additional resources are required to realize the idea,

the priority can be a guide for optimally allocating the

resources. Evaluation of the tree structure becomes

more accurate if nodes with more abstract content are

close to the root and nodes with more specific content

are closer to the leaf of the tree structure. This enables

a user who wants to know the outline to simply refer

to the node of the parent and a user who wants to

know specific examples and details to refer to the

child nodes of the tree structure. This tree structure is

considered suitable to describe well-organized ideas.

While there is currently no guarantee that the

excellence of the tree structure leads directly to the

excellence of ideas, we feel it is obvious that more

organized ideas can be found if the data is structured

appropriately.

Evaluation of the tree is currently being carried

out by analysis of the text in each node. First, the

evaluation method performs a morphological analysis

358

Takeshima, R. and Nagao, K..

Tool and Evaluation Method for Idea Creation Support.

In Proceedings of the 7th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2015) - Volume 1: KDIR, pages 358-363

ISBN: 978-989-758-158-8

on the text in the node, then it segments the text into

a word sequence, and finally it extracts its content

words. Each content word is converted into a vector

representation. This vectorization uses a machine

learning algorithm called word2vec (Mikolov et al.,

2013a) (Mikolov et al., 2013b). Wikipedia content

was used for the learning of word2vec. Using the

vectors of the word sequence, the method calculates

the semantic distance between the tree nodes and the

direction of abstraction based on the inner product of

the vectors. Each tree node is vectored as a mean

vector of the word vectors of its text content. If

semantic distances between the parent node and

several child nodes are closer, the tree structure is

better balanced, and if the inner product of the

difference vector of the parent node and its child node

and the difference vector of its child node and its

grandchild node is larger, the abstraction direction of

the tree is more semantically coherent. On the basis

of this vector calculation, we can determine the

relative advantages of the ideas in the tree structures.

Our idea creation support tool always calculates

the evaluation value of the idea and feeds the results

back to the user, thus helping the user create a more

sophisticated idea.

In this paper, first, we describe the support tool to

express an idea in a tree structure and the mechanism

to clearly visualize it by a poster. Next, as a means to

compare the ideas expressed in a tree structure, a

method of calculating the priority of the tree structure

is presented. Finally, in order to facilitate idea

creation in a group, we explain our meeting support

system, which presents a poster at a meeting and

enables the idea to be shared among all members. The

system aggregates the ideas evolved by each member

and integrates them into a larger tree structure.

2 IDEA CREATION SUPPORT

TOOL

2.1 Problem of Idea Creation

Idea creation is a central part of activities such as

research and product development. In idea creation, it

is important to know the entire content of the theme

being discussed, including peripherals such as

background and potential applications of ideas.

If we do not have an overall picture of the idea,

there is a danger of becoming too focused on the less

essential issues, which sometimes causes delays in

progress. If the idea creation is completed in a short

period, grasping the overall picture is relatively easy.

However, when idea creation requires a longer period

of time, it becomes difficult. In this latter case, it is

effective to hear the opinions of others in periodic

presentations. However, in a presentation, the focus

is often on the local problems of ideas, and it is

generally quite difficult to evaluate such a discussion

in a way that encompasses the orientation of overall

ideas.

2.2 Related Work

So far, there have been a few tools and methodologies

developed to support idea creation.

A typical example is the Mind Map (Buzan, T.,

1990) (Beel and Langer, 2011), which represents an

idea graphically by showing the relationship between

the elements of ideas in a tree structure. However, the

ability to create an intuitively understandable

expression depends on the author: the tool does not

automatically calculate the appropriateness of

representation and does not give any advice to

improve the content.

For collaborative idea creation, we typically

conduct meetings in order to develop the idea. There

are various systems in place to record the meeting

content in chronological order and allow members to

search and summarize the contents (Ishitoya, Ohira

and Nagao, 2009). However, we feel it is not

appropriate to use the time-series structure to

represent the content created during idea creation

activities because often several ideas are developed in

parallel and we sometimes reconsider previous ideas.

It is impossible to fully grasp the overall idea when

information is presented in sequence.

2.3 How to Support Idea Creation

Structurally describing the contents of an idea may

impose a large burden on the user, as it may entail not

just the description of notes and sketches but also an

awareness of the relationships between these.

Moreover, with the concept of the description, it starts

out as an abstract description and then the tree

structure typically divides into more specific

examples of the description. However, in the process

of refinement, another abstract description may be

derived, which may worsen the balance of the tree

structure. This makes it difficult to grasp the overall

picture. Therefore, in this work we focused on how

the presentation should be performed in order to best

explain the ideas to others.

In a typical presentation, visual aids such as slides

and posters are required. Therefore, we developed a

tool that can be used to create the tree structure of the

Tool and Evaluation Method for Idea Creation Support

359

idea easily and create a poster automatically. Figure 1

shows an example screen of the developed tool.

Posters created by this tool are called “digital posters”

and can include various contents such as text, images,

videos, and slides. Moreover, interactive operation of

the posters is possible, such as video playback and

slideshows.

Figure 1: Example screen of the developed tool.

The user can easily estimate the appropriateness

of the tree structure rather than simply creating the

tree structure data because the tree structure is

automatically converted into a poster by the system.

Moreover, since the user can directly use the outcome

of the idea creation in the presentation, the system

helps suppress any decrease in motivation for idea

creation.

2.4 Semantic Zooming

It is difficult to display the whole idea on one poster

as the tree structure of ideas is expanded and the

complexity of the idea increases. Therefore, the

system calculates the score of each node by using the

hierarchical relationships in the tree structure and the

attributes added to each node by the user. Then,

according to the score, the system presents the content

in a simplified manner. When the user wants to view

the nodes with the lower scores in more detail, he or

she just selects them in the poster and requests their

expansion. Expansion of the nodes does not just

magnify them visually—they are also semantically

refined in an operation called “semantic zooming”.

Semantic zooming displays the abbreviated content

of the selected node according to the size of the usable

area. An example of semantic zooming is shown in

Figure 2.

It is possible to arbitrarily change the degree of

omission according to the importance of the content

of each node, and it is also possible to adjust the

visibility of each node.

Figure 2: Semantic zooming.

Semantic zooming is available not only during a

presentation but also whenever the users who are

sharing a poster feel they want to.

3 IDEA EVALUATION METHOD

Sophisticated ideas can be expressed as a

semantically good balanced tree structure. The

semantically good balanced tree has parent and child

nodes that constitute an abstract/concrete relationship.

In other words, if the higher-level nodes of a tree

include abstract descriptions and the lower-level

nodes of the tree include concrete and detailed

description of the contents of their parent nodes, the

tree is semantically well-organized, i.e., the idea is

deeply constructed. While it sometimes happens that

interpretation of an abstract/concrete relationship

between two nodes changes by the context, the

meaning of that hierarchy will be well-defined if the

parent node has multiple child nodes. Therefore, it is

possible to calculate the semantic relevance of the

tree on the basis of the appropriateness of an

abstract/concrete relationship to a node having

multiple child nodes. It is also possible to evaluate the

quality of ideas on the basis of the evaluation of trees.

In this case, we consider only the node containing the

text and will find another mean for the nodes of the

images and video.

3.1 Conversion of Tree Node to Vector

To calculate the appropriateness of the

abstract/concrete relationship, our proposed method

converts the node to a vector representation that

enables numerical operation between nodes. For the

vector of the node, we utilize the vector

representation of the word, called “word2vec”

(Mikolov et al., 2013a) (Mikolov et al., 2013b).

word2vec uses a neural network called a “Skipgram”

KDIR 2015 - 7th International Conference on Knowledge Discovery and Information Retrieval

360

along with the result of the machine learning from

text corpora.

First, using word2vec, the method obtains vectors

of words contained in the node. We used Wikipedia

text for the learning of word2vec. The number of

dimensions of the vector to be learned was set to 200.

For the input and output of the Skipgram neural

network, we used the basic form of the noun and the

verb, respectively. This is due to the fact that nouns

and verbs represent the characteristics of the contents

of the node. An intermediate layer of Skipgram is the

word vectors of interest, and on the basis of this, the

method calculates the vector of the node, as follows.

The statements contained in the node are

morphologically analysed to extract the basic form of

the nouns and the verbs. Using word2vec, the method

obtains a vector representation of the extracted word.











∑









∈







w∈





(1)

By formula (1), the method calculates a mean

vector of words ( w ) contained in the

node( 



) and treats it as a vector for the node.

Regression analysis is performed using a set of

average vectors of nodes in the multiple layers of the

tree structure.

As a result, it discovers the axis of the concept of

the word meaning that represents the

abstract/concrete relationship of the tree structure.

The distance between the vectors of the nodes

represents the degree of the difference of meaning of

the descriptions of the nodes. As shown in Figure 3

and Figure 4, when three nodes A, B and C are close

to the axis of the abstract/concrete relationship, it can

be said that A has a description that is an abstract or

concrete version of the description of B and C. If node

D is separated by a greater distance from the axis, it

is considered that D has no strong abstract/concrete

relationship with A, B and C.

3.2 Priority of Idea

To determine superiority or inferiority among ideas,

it is judged whether the parent and child nodes of the

tree structure have an appropriate abstract/concrete

relationship. For a certain subtree, a preference in

terms of how appropriate the abstract/concrete

relationship is between the root node and its child

nodes is determined on the basis of the following

three evaluation values.

The preference of a subtree is higher if the average

error of the distance between the root node and its

child nodes is smaller. The distance between nodes

represents a difference of their

Figure 3: Tree structure of idea.

Figure 4: Abstract/concrete relationship.

meanings. For parent-child relationship between

nodes, when the distances between the parent

node and multiple child nodes are equal, it

indicates that the level of meaning between

child nodes is similar. In a subtree, the root node

A and its child nodes B, C, and D are like Figure

5, where nodes B and C have the same distance

from node A. This means B and C are more

appropriate as the child nodes of A. Meanwhile,

node D is separated by a greater distance from

node A in comparison with other child nodes.

This suggests that there is some problem with

connecting node D directly with node A.

• The preference of a subtree is higher if the sum

of the distances of the child nodes from the axis

of the abstract/concrete relationship of the

subtree is smaller and the child nodes are close

to the higher concreteness position of the axis.

When the distances between the parent node and

its child nodes are almost equal, if the child

nodes are located on the axis representing the

abstract/concrete relationship of the subtree, the

content of the parent node is considered an

abstraction of the content of all the child nodes.

In Figure 4, the distance between the parent

node A and the child nodes B, C, D are equal,

and nodes B and C are close to the axes

representing the abstract/concrete relationship,

so the content of node A is an abstract of the

contents of B and C. While the distance between

A and D is the same as that of the other child

Tool and Evaluation Method for Idea Creation Support

361

nodes, it is located further away from the axis

representing the abstract/concrete relationship.

Therefore, we assume that node D has

something wrong in its content or its position in

the tree.

• The preference of a subtree is higher if the

cosine similarity between the axis representing

the abstract/concrete relationship of subtree A

and the axis representing the abstract/concrete

relationship of subtree B whose root node is a

child node of A is larger. If the abstract/concrete

relationship between parent and child nodes of

the subtree and that between the child node and

the grandchild nodes is similar, it is considered

to have an appropriate abstract/concrete

relationship between the root node and its

grandchild nodes in the subtree.

Figure 5: Distance between nodes.

An evaluation value of the tree structure is

calculated for comprehensively evaluating the above

three features. When several ideas are being

considered, or when several alternatives are possible

to embody a certain idea, the tree structure of the idea

is evaluated to determine the relative advantages of

the alternative ideas by using this formula.

4 COLLABORATIVE IDEA

CREATION

4.1 Quotation of Ideas

Members of a group can share posters used at the

presentation through a common server. When the

poster is shared, its underlying tree structure of ideas

is also automatically shared, and then it becomes

quotable.

Each member has his or her own idea pool in which

they can store the idea(s) they have created. The ideas

that others have shared are stored in a separate pool.

A user can quote ideas by combining some of the

ideas of the others with his or her own. In this

Figure 6: Copy and paste of subtree.

operation, the entire or part of a tree structure is

copied and pasted as a child of a node of the other tree

structure as shown in Figure 6.

Two types of copy are possible: one, a quotation

of the idea itself (hot copy), and two, a duplication of

ideas that have been shared (cold copy).

With hot copy, when the original creator of the

ideas that are shared edits them, the editing result is

reflected in the copied nodes. However, a quoter

cannot edit the content of the copied nodes. Moreover,

the quoter is able to know when information has been

edited by the original creator and can decide whether

to reflect the result of that edit. After browsing the

results of editing, the quoter can cancel the change

and restore the content to that of before editing.

With cold copy, a quoter can edit the content of

the idea that was copied and quoted. It is possible to

notify the original creator when the quoter has edited

it. Then, the original creator can refer to the results of

editing and decide whether or not to modify the idea.

This mechanism of quoting ideas makes

collaborative idea creation by a group more feasible.

4.2 Synergy of Idea Creation

Idea creation by a group proceeds by the group

members sharing, quoting, and expanding ideas.

However, in many cases, members get caught up on

insignificant details, and the idea is likely to diverge.

By integrating and evaluating ideas that have evolved

separately, the proposed system attempts to converge

several ideas to create a more sophisticated one.

For this purpose, first, a tree structure is integrated

on the basis of quotation history. For hot copy, the

quoted results are shared when they are used in the

presentation. If the version of the quoted idea is not

changed, it is possible to integrate the idea

automatically. If the original idea has been edited, the

system integrates the original and quoted ideas after

some minor editing. In the case of cold copy, the

original idea has been edited, so the system cannot

perform a simple integration. The system splits the

KDIR 2015 - 7th International Conference on Knowledge Discovery and Information Retrieval

362

tree structure, or the original creator adopts a change,

and then it is possible to integrate.

After ideas are integrated, the resulting tree

structure can be very complex. The system therefore

uses the tree structure evaluation mechanism to

evaluate the tree and help users simplify the idea.

First, the system individually evaluates the

branched subtrees and identifies their relative

advantages. Also, during the presentation, the system

prompts users to vote on parts of the poster content

and assigns these additional evaluation values to the

nodes of the tree.

The system selects and adjusts the idea on the

basis of its evaluation of the tree structure and

presentation.

Idea selection is done by choosing the subtree that

has the highest value from among an assortment of

subtrees. Idea adjustment is done by correcting the

position or modifying the contents of the node as the

evaluation value of the tree structure becomes higher.

Our proposed idea creation support tool is

designed to help users perform these operations easily.

The system can hide the less important subtrees by

adding the attribute to their root nodes. The

evaluation value is recalculated after editing the

contents and relocating the nodes.

4.3 Task Management

Meetings for idea creation can include managing

tasks and keeping task of the progress required to

achieve the idea. In such cases, the tasks include

research to refine the idea and prototyping to embody

the idea.

The system inserts a node that represents the

content and the progress of the task to the node of the

corresponding idea. Each group member uses this

meeting support function to keep track of the

achievements and priority level of his/her tasks.

The degree of achievement depends on how much

refinement and implementation of the idea have been

achieved, and priority is quantified by whether the

task contributes to any part of the tree structure of

ideas (the node that is closer to the root node has a

higher priority).

The productivity of meetings is greatly improved

since the refinement of ideas and the task

management are carried out at the same time.

5 FUTURE PLAN

The primary focus of our future work will be the

operation and evaluation of the developed tool. On

the basis of operational experience with other tools

(Ishitoya, Ohira and Nagao, 2009), we will perform a

three-month trial and refine the tool. After that, we

will deploy the tool in a more ordinary situation for

more than half a year.

In particular, we will evaluate the effectiveness of

the evaluation of the idea by the tree structure. We

will explore whether the evaluation of the

abstract/concrete level in the hierarchy of the tree

structure contributes to a reasonable indication of

ideas that have been well-organized.

We innovate human creative activities by the

practical application of the idea creation support tool

that is the result of this study.

Detailed data on the various intellectual activities

of human beings collected by our system can be made

available by integrating an idea creation support tool

such as the Mind Map, a presentation tool such as

PowerPoint, or a task management tool such as

groupware. The collected data is a good guide for

considering the relationship between structures of

ideas and ease of understanding presentation, for

enabling idea integration by adjusting tree structures,

and for optimizing design and management methods

of tasks for the implementation of ideas.

In the near future, the accuracy of machine

learning will improve, which will lead to even better

support of idea creation and implementation.

REFERENCES

Buzan, T., 1990. Use Both Sides of Your Brain. Plume

Books.

Beel, J. and Langer, S., 2011. An Exploratory Analysis of

Mind Maps. In Proceedings of the 11th ACM

Symposium on Document Engineering, pp. 81-84.

Mikolov, T., Chen, K., Corrado, G. and Dean, J., 2013.

Efficient Estimation of Word Representations in Vector

Space. arXiv preprint, arXiv:1301.3781.

Mikolov, T., Sutskever, I., Chen, K., Corrado, G. S. and

Dean, J., 2013. Distributed Representations of Words

and Phrases and their Compositionality. In Burges, C.

J. C., Bottou, L., Welling, M., Ghahramani, Z. and

Weinberger, K. Q. (eds), Advances in Neural

Information Processing Systems, pp. 3111-3119.

Ishitoya, K., Ohira, S. and Nagao, K. 2009.

TimeMachineBoard: A Casual Meeting System

Capable of Reusing Previous Discussions. In

Proceedings of the Fifth International Conference on

Collaboration Technologies.

Tool and Evaluation Method for Idea Creation Support

363