Slide 1

Slide 1 text

Semantic Content Networks Next Step of Creating Topical Map

Slide 2

Slide 2 text

Basics - Index • Knowledge Base • Knowledge-based Trust • Semantic Networks • Frames • Frame Scripts • Concept Map • Topic Map • Concept Graph • FrameNet • Semantic Content Networks

Slide 3

Slide 3 text

Knowledge Base • Knowledge-base consists of the related existences, and their dimensions. • A knowledge-base is different from knowledge-graph. • Knowledge-graph is the graphical version of knowledge-base. • A knowledge base involves the facts, or misinformation with different labels. • A search engine defines every website as a “knowledge-base”. • A search engine has its own “knowledge-base” too. • So, if we have a knowledge-base, what is a semantic network? • Semantic Network is the different variations of connections from the same elements of the knowledge-base. • “X” can be defined 99 different phrase variations. • “X” can have 6 context domain connections, 2 Knowledge Domain connections, and 9 entities from 3 different entity types. • For all these connections, search engine can find different answers, or propositions as labeled different IDs. Programmable search engines Inventor Ramanathan V. Guha Assignee Google LLC

Slide 4

Slide 4 text

Knowledge Base • Semantic Network is the different variations of connections from the same elements of the knowledge- base. • “X” can be defined 99 different phrase variations. • “X” can have 6 context domain connections, 2 Knowledge Domain connections, and 9 entities from 3 different entity types. • For all these connections, search engine can find different answers, or propositions as labeled different IDs. Detecting spam search results for context processed search queries Inventor Ramanathan V. Guha Assignee Google LLC

Slide 5

Slide 5 text

Knowledge Base • To understand the Semantic Search Engines, “this year”, we will focus on the Ramanathan V. Guha. • “Detecting spam search results for context processed search queries” focus on using queries to understand the spam documents. • The “Product Line” and “Product Graph” are relevant here. • A “fake product name”, or a wrong match of “product-brand” association can make a web document “spammy”. • A knowledge base is necessary to prevent it. Detecting spam search results for context processed search queries Inventor Ramanathan V. Guha Current Assignee Google LLC

Slide 6

Slide 6 text

Knowledge Base • This one focuses on “aggregating context”, rather than processing it. • It tries to use a specific mechanism to define the knowledge of facts. • “Query Definitions, Aspects and Rephrasification” are relevant here. • A knowledge base can be used to understand overall context of a specific entity. • A search engine can filter pages based on their “context hints”, like design, structured data, relational layout elements, or currencies, statistics, measurement units, and brands, products, or other types of concepts. Aggregating context data for programmable search engines Inventor Ramanathan V. Guha Assignee Google LLC

Slide 7

Slide 7 text

Knowledge Base • A search engine, naturally, can classify the search results. • The thing here is that Ramanathan V. Guha focuses on a knowledge-base like “representation of things” to classify the documents. • There are many methods for document classification. • But, it categorizes query, if the ranking is good, it assumes that the document should be labeled with a category. • A knowledge base is necessary to classify the query, and query classification is necessary for document classification. Classifying search results Inventor Ramanathan V. Guha Assignee Google LLC

Slide 8

Slide 8 text

Knowledge Base • Search Result Ranking Based on Trust focuses on “factuality” and “accuracy” of a source. • It tries to rank documents based on the comprehensiveness. • Not a surprise to see “cancer” as an example on the patent. • It explains the trust signals. • It focuses on “vertical knowledge domain websites” for further trust signal information. • He tries to use words around the links for labeling them. • Negative reviews are thought to be used as well. Search result ranking based on trust Inventor Ramanathan V. GuhaCurrent Assignee Google LLC

Slide 9

Slide 9 text

Knowledge Base • “Query Identification” is different from “Query Aspects” and “Query Definitions”. • An early example of the “query categorization” and “query-category map”. • It uses the “query-page” matching from the “query logs” which signals the prominence of historical data. • The “QP Tuples” are the “Query-Page” Tuples. • It uses “query logs”, thus, if you are not in the query logs despite your quality, you can be overridden. • Knowledge Base Relevancy: It has to categorize the things on the queries and documents to increase the precision. Ramanathan V. Guha, Shivakumar Venkataraman, Vineet GuptaGokay, Baris Gultekin, Pradnya Karbhari, Abhinav Jalan Query identification and association Inventor Assignee Google LLC

Slide 10

Slide 10 text

Knowledge Base • Generating specialized search results just for a pattern of queries. • I focused on “question- query-document” templates last year. • So, I will skip it. • But, it uses “knowledge base” for understanding the data representation. Generating specialized search results in response to patterned queries Inventor Ramanathan V. Guha Current Assignee Google LLC

Slide 11

Slide 11 text

Knowledge Base • It is a good feeling to follow Mr. Ramanathan. • Because, he tries to find answers for the important questions. • “Corroboration of Web Answers”. • Co-identification of objects across heterogeneous information sources is prominent, because it provides a “cross-check” for the facts. • The ranking search results based on trust, and “co-identification of objects” should be handled together. • Query-augmentation is processed in this patent by “replacing predicates”. • Predicate is the heaviest term in the query. • Thus, “pattern matching” is relevant here too. Heuristic co-identification of objects across heterogeneous information sources Inventor Ramanathan V. Guha Current Assignee Google LLC

Slide 12

Slide 12 text

Knowledge Base • Providing search results is an important patent for the basics. • It explains the “entity types”, attributes, and HTML Code connections. • It focuses on the “templates” for specific situations. • The examples are for “restaurants”. • Thus, in the next slides, remember the concept, “frame script”. Providing search results Inventor Ramanathan V. Guha Current Assignee Google LLC

Slide 13

Slide 13 text

Knowledge Base • Most of the basic Google functions are processed by Ramanathan V. Guha. • This is important, because it makes a differentiation between “worthy result”, and “regular result”. • A worthy result is a signal for the quality. • Even, Google Alert, today, can be used to find some “worthy” results. Customized web summaries and alerts based on custom search engines Inventor Ramanathan V. Guha Current Assignee Google LLC

Slide 14

Slide 14 text

Who is Ramanathan V. Guha? Why do we focus on him? • He is important. • Creator of Schema.org. • Creator of RSS. • Creator of RDF. • He is a “Google Fellow” means “distinguished engineer”. • He has been cited in many books.

Slide 15

Slide 15 text

Who is Ramanathan V. Guha? Why do we focus on him? • Semantic Web, Making Sense, and other specific organizations or event cited him regularly. • He mainly focuses on “Contexts”.

Slide 16

Slide 16 text

Who is Ramanathan V. Guha? Why do we focus on him? • He has written a book. • He published many speeches about Google and Open Web. • At the right, you see his book, “Contexts – A formalization and some applications”. • And a screenshot from “Light at the end” presentation. • He explains why the “net” can’t be owned. • It is good to hear it from a Google Fellow.

Slide 17

Slide 17 text

Who is Ramanathan V. Guha? Why do we focus on him? • DataCommons is important from his speech. • DataCommons is a Knowledge Base. • Google integrated the DataCommons into their internal search system. • They defined the project as below.

Slide 18

Slide 18 text

Who is Ramanathan V. Guha? Why do we focus on him? • Since, he was able to organize four different search engines, it is a quite big success. • But, the main prominence of Ramanathan is that he thought using a “knowledge base” for every search engine problem. • Then, he tried to solve problem. • We can mention Justin Boyan and Data-highlighter tool, or Machine Learning/Rule-based System Transition on Google, as well.

Slide 19

Slide 19 text

Who is Ramanathan V. Guha? Why do we focus on him? • These are some of the entity types, and their possible relation types in a structured data sample. • Categories for occurrence, or categories by domains represent how these things help search engines to structure their own “semantic networks” and knowledge bases.

Slide 20

Slide 20 text

Who is Ramanathan V. Guha? Why do we focus on him? • Ramanathan also worked with the web masters. • He visualized the structured data as a form of “semantic network”. • Thus, from a Knowledge Base Concept to the Relational Databases, he has a prominent place. • He created some guides for SEOs too.

Slide 21

Slide 21 text

Who is Ramanathan V. Guha? Why do we focus on him? • Vocabulary represents the richness of the content in terms of unique words. • But, Ramanathan uses it for different topic grasping algorithms. • On his talk, during 2014, he mentioned Scholarly Works, Comics, Serials, and Sports. • After that, Schema.org started to launch new SD Types.

Slide 22

Slide 22 text

Who is Ramanathan V. Guha? Why do we focus on him? • Vocabulary represents the richness of the content in terms of unique words. • But, Ramanathan uses it for different topic grasping algorithms. • On his talk, during 2014, he mentioned Scholarly Works, Comics, Serials, and Sports. • After that, Schema.org started to launch new SD Types.

Slide 23

Slide 23 text

Who is Ramanathan V. Guha? Why do we focus on him? Aggregated Knowledge Graph => Does it sound similar from the patents?

Slide 24

Slide 24 text

Who is Ramanathan V. Guha? Why do we focus on him? Aggregated Knowledge Graph => Does it sound similar from the patents?

Slide 25

Slide 25 text

Knowledge Base • This is another prominent sample that I preferred to put with some common inventors. • “Alternatives of an answer, and a question” represent the “slight differences on wording”. • And, it requires to have a “knowledge base”, since it focuses on Information Extraction. • Understand the difference between “Occurrence Tracking Search Engine”, and “Perception based Search Engine”. Determining question and answer alternatives Inventor David Smith, Engin Cinar Sahin, George Andrei Mihaila Current Assignee Google LLC

Slide 26

Slide 26 text

Knowledge Base • From the same fellows, you have a “Context”. • They cite the Ramanathan in the invention. • They mention the “Topics” and “Movie/Book” intersection of the Harry Potter. Determining question and answer alternatives Inventor David Smith, Engin Cinar Sahin, George Andrei Mihaila Current Assignee Google LLC

Slide 27

Slide 27 text

Knowledge-based Trust • Knowledge-based Trust involves a PageRank balancing effect via Accurate, Unique and Comprehensive Information on sources. • “Ranking search results with Trust” via a Knowledge Based evolved into “Knowledge-based Trust” in the same year. Explained Before

Slide 28

Slide 28 text

Knowledge-based Trust • Knowledge-based Trust involves a PageRank balancing effect via Accurate, Unique and Comprehensive Information on sources. • A single “broken sentence” can decrease the granularity of the semantic network that will be extracted. Explained Before

Slide 29

Slide 29 text

Knowledge-based Trust • “Extractor”, or “Constructor”… • It is a representation of the trust-measurement framework. • The “Directed Graph” model explains how the “value pairs” are being extracted. Explained Before

Slide 30

Slide 30 text

Knowledge-based Trust • “Source Accuracy” represent the trust that a search engine can give for visibility. • The algorithm design simply focus on two main steps. • Extracting the triples. • Comparing them to the others, especially the ones in Knowledge Base. Explained Before

Slide 31

Slide 31 text

Knowledge-based Trust • An example of the “Triple Correctnes Prediction”. • “Freebase Triple” is important. • Freebase is bought by Google to forge the Knowledge Graph. • The “triple” from Knowledge Base is compared to the triples from the SERP. Explained Before

Slide 32

Slide 32 text

Knowledge-based Trust • Seed Sources and Tail Sources. • Remember the RankMerge. • Remember the Topic- sensitive PageRank. • KBT is not an alternative of PageRank. • It is a Hybrid Search Engine from Hypertextual to Semantic. Explained Before

Slide 33

Slide 33 text

Semantic Networks • Semantic Networks are the knowledge representations with the relational databases. • A knowledge base can have multiple semantic networks from the same knowledge base dimensions. • All the knowledge representation schemes should be human- understandable. • Semantic Networks and Frame Representations are connected to each other.

Slide 34

Slide 34 text

Semantic Networks • Knowledge Bases represent the Knowledge. • Thus, Knowledge Representation is a central term. • Logical Representation is “difficult for common sense”. • Logical Representation has uncertainty, and it has time constraints, along with beliefs. • Logic representations require “reasoning”. Basically, it is “TrueVTrue = True”. • These rules are not good for extracting the facts, and storing them in a relational database. Entity ID of m076hq

Slide 35

Slide 35 text

Semantic Networks • Semantic Networks represent factual knowledge in classes of objects and their properties. • Declarative knowledge representations, and their relations are stored. • Static representations (knowledge that stays same) are more common for semantic networks. • Semantic Networks imitate human memory. • Structured Object Representation include properties, nodes and edges within Directed Graphs. • Thus, it is called “Association Graph”. site:https://artsandculture.google.com/entity/

Slide 36

Slide 36 text

Semantic Networks • General Networks shouldn’t be considered as Semantic Network. • Every “arch”, knowledge representation has to have a relation type. This is a general network.

Slide 37

Slide 37 text

Semantic Content Networks • We need “relation types”.

Slide 38

Slide 38 text

Semantic Content Networks • Lexical Parts of a Semantic Content Network: • Nodes – objects from different classes, or individuals. • Links – relationships between the objects. • Labels – nodes and links, together. • Procedural Part: • Constructors • Destructors • Writers • Readers Structural Part has the “links and nodes”. Semantic part has the “associated meanings” based on the knowledge domain.

Slide 39

Slide 39 text

Semantic Content Networks • Relation Type Examples: And/OR Trees This is relevant to “Boolean Retrieval”.

Slide 40

Slide 40 text

Semantic Content Networks • Relation Type Examples: IS/A Hierarchy Is/A relation type is relevant to the Taxonomy, Lexical Relations, Semantic-Dependency Tree, and Augmented Information Retrieval. All these are relevant to Set Theory.

Slide 41

Slide 41 text

Semantic Content Networks • Relation Type Examples: IS/Part Hierarchy Is/A relation type is relevant to the Taxonomy, Lexical Relations, Semantic-Dependency Tree, and Augmented Information Retrieval. All these are relevant to Set Theory. Lexical Relations help an SEO to focus on meaning, rather than query search demand (keyword volume.)

Slide 42

Slide 42 text

Semantic Networks 1. Hypernym: The general word of another word. For example, the word color is the hypernym of red, blue, yellow. 2. Hyponym: The specific word of another general word. For example, crimson, violet, lavender are the hyponyms of purple. And, purple is the hyponym of the color. 3. Antonym: The opposite of another word. For example, the big is the antonym of the small, and the early is the antonym of the late. 4. Synonym: The replacement of another word without changing the meaning. For example, huge is the synonym of big, and initial is the synonym of early. 5. Holonym: The whole of a part. For example, the table is the holonym of the table leg. 6. Meronym: The part of a entire. For example, a feather is the meronym of a bird. 7. Polysemy: The word with different meanings such as love, as a verb, and as a noun. 8. Homonymy: The word with different meanings accidentally, such as bear as an animal and verb, or bank as a river or financial organization.

Slide 43

Slide 43 text

Semantic Networks Inheritance helps search engines to Extract Information by augmenting. Semantic Dependency Tree is a representation of inheritance. Source: Ryan J. Urbanowicz.

Slide 44

Slide 44 text

Semantic Networks Every arch defines a binary relation. For the sentence, “John’s mother has sued her husband at the age of 42.” The relations here are “predicate”, “age”, and “marriage status”. Source: Ryan J. Urbanowicz.

Slide 45

Slide 45 text

Semantic Networks Arch Types: X is a Y. => Individual to Concept => X is a tiger. X is a kind of Y. => Concept to Concept => X is a wild kind of animal. X is related to Y. => Individual to Individual => X Tiger leads the pack of Y Tigers. Source: Ryan J. Urbanowicz.

Slide 46

Slide 46 text

Semantic Networks Reification: Relationships can be turned into a frame. A frame can turn non-binary relationships into an object. A giver A recipient An object Source: Ryan J. Urbanowicz. Learn, Semantic Role Labels.

Slide 47

Slide 47 text

Semantic Networks Multiple Inheritance Source: Ryan J. Urbanowicz. Learn, Semantic Role Labels.

Slide 48

Slide 48 text

Semantic Networks Multiple Inheritance - Range Queries/Range Values Source: Ryan J. Urbanowicz. Learn, Semantic Role Labels.

Slide 49

Slide 49 text

Semantic Networks Intersection Search Source: Ryan J. Urbanowicz. Learn, Semantic Role Labels.

Slide 50

Slide 50 text

Semantic Networks Complex Semantic Networks Source: Ryan J. Urbanowicz. Learn, Semantic Role Labels.

Slide 51

Slide 51 text

Semantic Networks Inferential Distance Source: Ryan J. Urbanowicz. Learn, Semantic Role Labels.

Slide 52

Slide 52 text

Semantic Networks Partial Ordering – How the semantic network should be shaped? Source: Ryan J. Urbanowicz. Learn, Semantic Role Labels.

Slide 53

Slide 53 text

Semantic Networks An example: From Hyper-structured to Structured Data: Source: Ryan J. Urbanowicz. Learn, Semantic Role Labels.

Slide 54

Slide 54 text

Semantic Networks An example: From Hyper-structured to Structured Data: Source: Ryan J. Urbanowicz. Learn, Semantic Similarity.

Slide 55

Slide 55 text

Semantic Networks An example: From Hyper-structured to Structured Data: Source: Ryan J. Urbanowicz. Learn, Semantic Role Labels.

Slide 56

Slide 56 text

Semantic Networks An example: From Hyper-structured to Structured Data: Source: Ryan J. Urbanowicz. Learn, Semantic Role Labels.

Slide 57

Slide 57 text

Semantic Networks Unified Medical Language System is an Example of Semantic Network. Source: Ryan J. Urbanowicz. Learn, Semantic Role Labels.

Slide 58

Slide 58 text

Semantic Networks A website can make the semantic network of the search engine richer and more accurate. It can be the authority of truth, not just the topic.

Slide 59

Slide 59 text

Semantic Networks An example of Semantic Network Visualization.

Slide 60

Slide 60 text

Semantic Networks An example of Context Vectors.

Slide 61

Slide 61 text

Semantic Networks Frames: Representation of Stereotypes. Frames have inheritance. Frames can have different data structures. Frames are the typical semantic networks for the same types of things. Frames are useful for training AI. At the right, you see a Frame. Every class, and sub-class have specific attributes. All the relation types are clear. In this case, Knowledge Bases are more similar to the Frames. Frames have “slots”, and “filters to be filled in. Values can be static knowledge. A “daemon” can exceed the threshold and make the knowledge “dynamic”.

Slide 62

Slide 62 text

Semantic Networks Semantic Network Frame

Slide 63

Slide 63 text

Semantic Networks Every frame slot can have a slot. Every slot can have a “value”, “default”, “range”, “if- added”, and “if-needed” property. Frames help an SEO to create specific types of document templates for specific types of query templates. Facets can include different types of semantic networks. • Represents the “default”. • Every facet here can be entirely different semantic network. • A central entity will reflect the core of the specific entity’s identity.

Slide 64

Slide 64 text

Semantic Networks • Situational Frame: Valid only for a situation. • Action Frame: Actions for the situation. • Combination Frame: Combination of situational and action frames. • General Frame: Hierarchy included class models. • These frames are for typical situations, the unique or rare entities might not be put into a frame. • At the right side, you will see another frame, but it is only for the restaurants. • In the next slide, you will see an example of FrameNet.

Slide 65

Slide 65 text

Semantic Networks • “Jack went to a restaurant. He decided to order steak. He sat there and waited for a long time. Finally, he got angry and left. • “What was Jack waiting for?” • “Why did he get angry?” • A Frame Script can shape a knowledge base for different situations. • Same script can extract all the information for the similar situations. • Remember the “Restaurant” sample from Ramanathan V. Guha.

Slide 66

Slide 66 text

Semantic Networks • Concept Maps: To create a Semantic Network, a concept map is necessary. • Every existence is connected to at least one concept. • “Eagle” can be connected to “flying”, or “hunting” concept.

Slide 67

Slide 67 text

Semantic Networks • Topic Maps: It sounds similar? • It reflects any kind of concept, existing things, or hypergraphs for associations, occurrences and real-world things. • A topic map connects concepts to the each other along with the real world entities. A topic map can be extracted from a content network, and turned into a semantic network. • This brings us to the “Concept Graphs”. • A concept graph is again for patterns. • At the right side, you will see an example of concept graph.

Slide 68

Slide 68 text

Semantic Networks • A FrameNet is different than a frame. • A FrameNet includes the semantic role labels but in the form of more detailed patterns. • For example, the sentence “The Unit found two men hiding in the trunk of a car…” includes a “sought entity”, “perceiver” and “location”. • If we use the sentence like “The Unit found three thiefs that are hiding in the storage”, relations are same. • It is similar to the “Frame Scripts”. • But, FrameNet is actually a programming output to help machines to read the semantic knowledge.

Slide 69

Slide 69 text

Semantic Content Networks • Semantic Content Networks are the next steps that come after the Topical Map. • A Semantic Content Networks contains a Context Vector, Context Hierarchy, Structure and Connection. • A Semantic Content Network is the textual and visual form of a semantic network. • A semantic content network can help a search engine to change the truth in the knowledge base. • A search engine shapes the SERPs based on the knowledge-base. • A semantic content network can have a high- level of knowledge-based trust, and search engine can start to shape the SERPs network based on the specific source. • Semantic content networks can’t be imitated. • A semantic content network can be beaten by another one. • A better context vector, hierarchy, connection, and structure are must.

Slide 70

Slide 70 text

Semantic Content Network Results

Slide 71

Slide 71 text

Semantic Content Network Results

Slide 72

Slide 72 text

Semantic Content Network Results https://www.oncrawl.com/technical-seo/creating-semantic- content-networks-with-query-document-templates-case-study/

Slide 73

Slide 73 text

Semantic Content Network Results https://www.holisticseo.digital/theoretical-seo/ranking

Slide 74

Slide 74 text

Semantic Content Network Results https://www.authoritas.com/blog/encazip-case-study Aggregator Energy Industry Encazip.com

Slide 75

Slide 75 text

Semantic Content Network Results Book Store Huge Success 5th Largest E-commerce store Of Turkey BKMKitap.com

Slide 76

Slide 76 text

Semantic Content Network Results Blackhat + Semantic SEO Expired Domain Drinks

Slide 77

Slide 77 text

Semantic Content Network Results Dog-food Industry Affiliate

Slide 78

Slide 78 text

Semantic Content Network Results Monitor Industry Affiliate

Slide 79

Slide 79 text

Semantic Content Network Results Kanban Industry Trained and SCN Designed

Slide 80

Slide 80 text

Semantic Content Network Results Blackhat + Semantic SEO Expired Domain

Slide 81

Slide 81 text

Semantic Content Network Results SaSS It was hard to train these people.

Slide 82

Slide 82 text

Semantic Content Network Results Re-ranking Triggered Dog-food Industry

Slide 83

Slide 83 text

Semantic Content Network Results Gift Industry New Launch Brand-new Domain 3th week

Slide 84

Slide 84 text

Semantic Content Network Results Exact Match Domain Car Industry 7th week

Slide 85

Slide 85 text

Semantic Content Network Results Affiliate Aggregator

Slide 86

Slide 86 text

THANK YOU