Menu Close

Use Cases

Five Possible Levels of Analysis of a Knowledge Graph

Aug 19 2022 

Level of Analysis

Tasks

Descriptions

Possible Methods

Element

(Concept or Event)

Information Retrieval

Retrieval of information about an element, such as its definitions, associated publications, authors, research models, or informing theories

Graph query language (GQL)

Element Semantic Similarity

The similarity between two elements based on their element labels or definitions

Bidirectional Encoder Representations from Transformers (BERT) (Devlin et al., 2018)

Jingle and Jangle Fallacies

The jingle fallacy occurs when two concepts have identical or similar labels but reference different real-world phenomena (Thorndike, 1913). The jangle fallacy occurs when two concepts reference identical or similar phenomena but are labeled differently (Kelley, 1927).

NLP-based similarity matrix using BERT for all definitions (Song, Watson, Zhao, et al., 2021)

Theme Identification

Identification of research themes based on concepts

Using concepts or their definitions as vectors and use of K-means or NLP for clustering

Element-topic correlation analysis

The association of elements and research topics

Correspondence analysis

Element Relationship

Information Retrieval

Retrieval of related elements, such as their antecedents, consequents, or all paths starting or ending with a concept or between two elements

GQL to retrieve connected elements. Graph theory methods such as shortest path analysis

Network Importance

The importance of an element in a knowledge network

Social network analysis metrics, such as centrality measures

Model

Jungle conundrum

Identification of similar causal models in a knowledge network (Song, Watson, Zhao, et al., 2021)

Graph isomorphism analysis (Song, Watson, Zhao, et al., 2021)

Endogeneity issues

Identification of potential endogeneity issues, such as omitted variables and simultaneity issues

DAG analysis methods (Textor et al., 2011), such as identifying direct paths, backdoor paths, and colliders

Critical elements

Identification of essential elements in a knowledge network

Social network methods, such as structural cohesion analysis to identify the critical nodes that connect a model or bring together different models.

Theory

Identification of theories

Retrieve theories

GQL

Theory impact

Analysis of the impact of a theory in a knowledge network

GQL to report the frequency of a theory appearing in a knowledge network

Theme

Knowledge fragmentation

Evaluation of the cumulative nature of a knowledge network

Social network density to compare networks

Literature gap

Identification of gaps in the literature

Social network analysis to identify structural holes and weak ties

Comparative analysis

Tracking the development of a theme by journal or time

Cluster analysis (Shmueli et al., 2017) to group concepts, definitions, and theories to identify themes in a set of journals or over time

Model integration or simplification

Condense and simplify causal models related to a particular topic area

Graph summarization methods, such as clustering, classification, pattern set mining, and outlier detection (Liu et al., 2018)

Examples of Knowledge Analytics

This appendix illustrates possible analyses of digitized knowledge, i.e., potential use cases of the search and analysis app (Analysaurus).

Querying is based on openCypher (Francis et al. 2018), an industry standard for querying graph databases[1], and R is used for further processing of query results.

Element Level: At the element level, we can measure the semantic similarity of elements using NLP methods. For example, Table A.1 shows three different semantic similarity measures of concepts based on their labels (Song et al. 2021b).

Concept 1 Concept 2 String-based(Cosine; Jaccard) Corpus-based(Latent semantic analysis) Knowledge-based (Wordnet)
Internet privacy concerns User privacy concerns 0.667 0.999 1
Information privacy concerns Privacy concern 0.408 0.998 1
Risks Costs 0 0.998 0.167
Trust Work overload 0 0 0.133
Costs Benefits 0 0.615 0.308
Table A.1: Semantic Similarity of Concepts based on Their Labels

In a body of literature, it is quite common to have multiple definitions for the same concept. However, when knowledge is digitized, concept similarity computation is possible (see Table A.2 for the example of trust). 

The similarity of concept definitions can be computed using NLP tools such as BERT. Inspection of Table A.3 reveals that definition 3 has the highest average similarity score and the most similar definitions are 1 and 5 (bold), while the least similar are 1 and 7, and 5 and 7 (italics).

  Definition Publication
1 We define trust as the subjective assessment of one party that another party will perform a particular transaction according to his or her confident expectations, in an environment characterized by uncertainty. Ba, S., & Pavlou, P. A. (2002). Evidence of the Effect of Trust Building Technology in Electronic Markets: Price Premiums and Buyer Behavior. MIS Quarterly, 26(3), 243.
2 Trust is argued to be rooted in perceptions of teammates’ ability, benevolence, and integrity (Jarvenpaa et al. 1998). Ability refers to the aptitude and skills that enable an individual to be perceived as competent by teammates (Jarvenpaa et al. 1998; Mayer et al. 1995). Benevolence refers to the extent to which an individual is believed to be willing to help teammates beyond personal motives or individual gain. 1995). Integrity refers to the extent to which an individual is believed to adhere to a set of principles thought to make her dependable and reliable. Piccoli, & Ives. (2003). Trust and the Unintended Effects of Behavior Control in Virtual Teams. MIS Quarterly, 27(3), 365.
3 Trust is defined as the buyer’s intentions to accept vulnerability based on her beliefs that the transaction will meet her confident expectations. Pavlou, Liang, & Xue. (2007). Understanding and Mitigating Uncertainty in Online Exchange Relationships: A Principal-Agent Perspective. MIS Quarterly, 31(1), 105. 
4 The user beliefs in the recommendation agents’ competence, benevolence, and integrity. The beliefs that 1) the recommendation agent has the ability, skills, and expertise to perform effectively 2) the recommendation agent cares about the user and acts in the user’s interest 3) the recommendation agent adheres to a set of principles (e.g., honesty and promise keeping) that the user finds acceptable, Xiao, & Benbasat. (2007). E-Commerce Product Recommendation Agents: Use, Characteristics, and Impact. MIS Quarterly, 31(1), 137. 
5 Trust reflects one party’s belief that its requirements will be fulfilled through future actions undertaken by the other party. Goo, Kishore, Rao, & Nam. (2009). The Role of Service Level Agreements in Relational Management of Information Technology Outsourcing: An Empirical Study. MIS Quarterly, 33(1), 119. 
6 Trust is conceptualized as a single variable and refers to general confidence in the website. Cyr, Head, Larios, & Pan. (2009). Exploring Human Images in Website Design: A Multi-Method Approach. MIS Quarterly, 33(3), 539. 
7 The extent to which a buyer perceives in a seller’s ability (i.e., skills, competencies, and characteristics in seller his/her products online), integrity (adhering to a set of principles that the buyer finds acceptable), and benevolence (i.e., doing good toward the buyer). Ou, C. X., Pavlou, P. A., & Davison, R. M. (2014). Swift Guanxi in Online Marketplaces: The Role of Computer-Mediated Communication Technologies. MIS Quarterly, 38(1), 209–230.
Table A.2: Definitions and Defining Publications for Trust

 

Definition 1 2 3 4 5 6 7
1
2 0.60
3 0.80 0.63
4 0.57 0.76 0.63
5 0.81 0.61 0.77 0.57
6 0.59 0.50 0.57 0.49 0.46
7 0.38 0.57 0.52 0.76 0.38 0.45
Means 0.62 0.61 0.66 0.63 0.60 0.51 0.58
Table A.3: Concept Similarity for Trust Definitions

Element-Relationship Level: Visualizations can be generated to focus on a particular element and its role. For example, the antecedents of trust can be seen in Figure A.1.

Figure A.1: Antecedents of Trust

All paths starting with an element, ending with an element, or between two elements can be detected (see Table A.4). 

Trust -> Perceived information asymmetry -> Perceived uncertainty -> Purchase intentions -> Actual purchases
Trust -> Information security concerns -> Perceived uncertainty -> Purchase intentions -> Actual purchases
Effective use of feedback system -> Presence -> Trust
Effective use of instant messenger -> Interactivity -> Trust
Satisfaction -> User control -> Trust
Behavior control -> Vigilance -> Trust
Incongruence -> Salience -> Trust
Table A.4: Causal Paths Starting or Ending with Trust

Social network analysis can holistically assess the importance of an element in a knowledge network (see Table A.5). Specifically:

A high Betweenness centrality suggests that an element is an influential mediator.

A high Closeness centrality aggregates the most direct causal influence that an element has on other elements in a knowledge network. 

A high Outdegree centrality of an element in a knowledge network indicates that an element is a fundamental antecedent. 

A high Indegree centrality score signals that an element is a commonly accepted consequent. 

Concept Betweenness centrality Closeness centrality Outdegree centrality Indegree centrality
Trust 726 0.38 17 61
Perceived uncertainty 126 0.26 2 4
Mutual dependence 81 0.28 28 16
Product expertise 65 0.31 9 9
Purchase intentions 44 0.21 1 2
Harmonious conflict resolution 40 0.28 15 10
Perceived information asymmetry 38 0.34 1 3
Fears of seller opportunism 38 0.34 1 3
Information privacy concerns 38 0.29 1 3
Information security concerns 38 0.29 1 3
Table A.5: Concept Network Centrality Measures

Model Level: Similarity can be computed for any pair of causal models for a domain of interest. Figure A.2 shows two privacy models and their conceptual isomorphism score, a measure between 0 and 1 (Song et al. 2021b).

Figure A.2: Similar Models for Privacy (Song et al. 2021a)

Theory Level: Theories guide knowledge creation. For the MISQ trust curation (Söllner et al. 2016), there is a wide variety of theoretical foundations, which could suggest that there is little coalescence around theories for explaining trust formation (see Table A.6).

Theory Frequency
Agency theory 3
Technology acceptance model 2
Social exchange theory 1
Social capital theory 1
Theory of planned behavior 1
Theory of reasoned action 1
Social presence theory 1
Theories of human information processing 1
The theory of interpersonal similarity 1
The theories of trust formation 1
The theories of satisfaction. 1
Innovation diffusion theory 1
Incomplete contract theory 1
Relational exchange theory 1
Visual rhetoric theory 1
Message exchange theory 1
Adaptive structuration theory 1
Media synchronicity theory 1
Table A.6: Theories Referenced in the MISQ Trust Curation