Methods<\/h2>\n
A list of URLs containing lists of book recommendations was analyzed to gather unique titles of books, as determined by ISBN-13 (1) identifiers. These URL sources for these recommendations were collected through a series of Google searches using keyword phrases related to an initial seed keyword phrase.<\/p>\n
Related keywords were prioritized in descending order of estimated monthly search volume in the US-Google database. Information related to related keywords and search volumes was provided by the SEMRush service (2). The analysis of URLs with book recommendation lists is detailed below as a series of six unique phases of analysis and quantification.<\/p>\n
Phases:<\/h3>\n\n- Keyword Discovery<\/li>\n
- Keyword prioritization<\/li>\n
- Relevant URL identification<\/li>\n
- Book Title Discovery<\/li>\n
- Book Title Filtering\/Exclusions<\/li>\n
- Book Title Frequency Sorting<\/li>\n<\/ol>\n
Keyword Discovery<\/h3>\n
The SEMRush marketing platform was used to identify target keywords related to recommended book titles on machine learning and artificial intelligence. An initial keyword search using the phrase “best machine learning books” was searched in the US database of the Google search engine.<\/p>\n
The URLs from the top 5 organic results of this search were obtained as seed URLs to gather related keyword terms to broaden search parameters for later searches. The SEMRush marketing platform was then used to gather search data related to total organic keywords and estimated monthly search traffic for these URLs.<\/p>\n
Filtering was applied to the organic position reporting data for each URL to ensure maximal relevancy. A positional filter (P) was used to eliminate keywords where the URL did not appear in an organic result at or below the 20th position. A volume-based sorting filter (V) was also applied to sort results based on the largest estimated monthly search interest. The results obtained from the initial search are reported in Table 1 and Table 2.<\/p>\n
Keyword Prioritization<\/h3>\n
The organic keyword results for each URL were filtered for those where the pages appeared within the top 20 results. This filter was applied to remove unrelated keyword phrases from reporting data.<\/p>\n
The organic keyword results were then sorted in descending order of volume (V) to identify the phrases that represented the largest numbers of monthly searches\u2014a proxy for interest. Results of this initial search can be found in Table 2.<\/p>\n
Relevant URL Identification<\/h3>\n
The top 10 keywords resulting from the keyword prioritization phase were used as seed keywords to conduct a series of Google searches. For each keyword among these 10, a unique Google search in the US database was performed to produce a unique search engine results page (SERP) from which the resulting top 5 organic page URLs were recorded. This process produced a set of 33 unique URLs.<\/p>\n
Book Title Discovery<\/h3>\n
The 33 unique URLs were then manually inspected to assess relevancy for containing book titles. Of the 33 URLs discovered only 16 were deemed relevant to book title discovery. URLs were disqualified via a manual inspection process on the grounds of either being relevant to only a single book title or not mentioning book titles. This represented a ~49% relevancy rate among URL inclusions.<\/p>\n
Relevant URLs were then manually inspected for book titles and any titles found included were added to a global list of unique titles, up to a maximum of 10 titles per URL source<\/strong>. This upper limit was put into place after observing several URLs to contain excessive numbers of relative titles. Where titles were found to be uniquely identified only by edition, all but the most recent edition were excluded. Another series of filtering for uniqueness was done by ISBN comparison when that data was available. Amazon.com was used as the source for ISBN listings.<\/p>\nExclusion Criteria<\/h3>\n
Given the statistical nature of this study, it was deemed appropriate to include as many recommendations from as many sources as practical. As such, many titles that were regarded as possibly irrelevant were included for consideration. The frequency of occurrence among several sources was maintained as an effective filter for excluding such titles.<\/p>\n
Titles without an ISBN number were excluded. In such cases, these entries were considered non-entries and an additional entry was considered from the list where available. There was only one instance in which this exclusionary filter was applied to a book titled Monroe Doctrine II<\/em>. This book was categorized as fiction.<\/p>\nIn the case of Amazon best-seller category lists, some titles appeared multiple times within the top 10 results as multiple formats. For example, the title “Life 3.0: Being Human in the Age of Artificial Intelligence” appeared both in the Kindle edition and Paperback edition. In this case, as in the case of similar duplicates, the physical version was selected and an additional entry was taken from the list.<\/p>\n
An additional series of exclusions were made after the titles were collected and frequency analysis was performed. This exclusion removed book titles appearing on fewer than 3 URLs. This filtering was done after noting that, in addition to 61 titles appearing on single URLs only, there were ample titles appearing on 3 or more sources to compile an adequate list.<\/p>\n
Title Frequency Sorting<\/h3>\n
The results from the title gathering process were then analyzed for frequency of mention among all considered recommending URLs. The list of book titles and corresponding frequency counts were then sorted in descending order to provide a list of the most frequent titles among all URLs recommending books. This ordered data then served as our basis for the list of best books on machine learning and artificial intelligence.<\/p>\n
Results<\/h2>\n
The total organic keywords after having the positional filter applied (P) and the total estimated monthly traffic of the top 5 organic URLs returned for the key term phrase “best machine learning books” are reported in Table 1.<\/p>\n
These results reflect known traffic distribution patterns of Google search results such that higher results receive roughly 35% of total clicks resulting in greater estimated monthly traffic counts.<\/p>\n![\"table](\"https:\/\/www.alpharithms.com\/wp-content\/uploads\/963\/table-1-initial-reserach-best-ml-books-1024x576.jpg\" "\\"table-1-initial-reserach-best-ml-books\\"")
<\/a>Table 1<\/strong>: Organic Keyword Counts & Search Volumes of Initial Search (click to enlarge)<\/figcaption><\/figure>\nAfter the collection of organic keyword data for each of these initial 5 URLs, a set of unique keywords found among all URLs was collected. This set was then sorted based on total monthly volume to, again, serve as a proxy representing cumulative interest. There were no filtering criteria applied for inclusion other than the appearance on any of the 5 URLs.<\/p>\n
This collection was then sorted in by monthly search volume in descending order. The keywords reflecting the 10 highest monthly search volumes were then selected as the keyword phrases URL discovery in the book title gathering phase. These results are reported in Table 2 below. A JSON-format file containing the collection of all keywords, their relative counts among the considered URLs, and the monthly volumes are available here for download<\/a>.<\/p>\n![\"table](\"https:\/\/www.alpharithms.com\/wp-content\/uploads\/963\/table-2-seed-keywords-and-monthly-volume-1024x576.jpg\" "\\"table-2-seed-keywords-and-monthly-volume\\"")
<\/a>Table 2<\/strong>: Observed frequency and reported monthly search volume of keywords in top 5 results of the initial search. (click to enlarge)<\/figcaption><\/figure>\nOf the unique URLs found within the SERP for these terms, a total of 148 book titles were discovered resulting in 89 unique titles for consideration. Of these 89 titles, only 28 appeared on more than one URL. This number was deemed impractical for representing a general consensus among considered URLs such that only book titles appearing on 3 or more distinct URL lists were included. A breakdown of these results is illustrated in Table 3.<\/p>\n![\"table](\"https:\/\/www.alpharithms.com\/wp-content\/uploads\/963\/table-3-book-title-frequency-chart-1024x576.jpg\" "\\"table-3-book-title-frequency-chart\\"")
<\/a>Table 3<\/strong>: Frequency distribution of book titles among considered URLs (click to enlarge)<\/figcaption><\/figure>\n
Keyword Discovery<\/h3>\n
The SEMRush marketing platform was used to identify target keywords related to recommended book titles on machine learning and artificial intelligence. An initial keyword search using the phrase “best machine learning books” was searched in the US database of the Google search engine.<\/p>\n
The URLs from the top 5 organic results of this search were obtained as seed URLs to gather related keyword terms to broaden search parameters for later searches. The SEMRush marketing platform was then used to gather search data related to total organic keywords and estimated monthly search traffic for these URLs.<\/p>\n
Filtering was applied to the organic position reporting data for each URL to ensure maximal relevancy. A positional filter (P) was used to eliminate keywords where the URL did not appear in an organic result at or below the 20th position. A volume-based sorting filter (V) was also applied to sort results based on the largest estimated monthly search interest. The results obtained from the initial search are reported in Table 1 and Table 2.<\/p>\n
Keyword Prioritization<\/h3>\n
The organic keyword results for each URL were filtered for those where the pages appeared within the top 20 results. This filter was applied to remove unrelated keyword phrases from reporting data.<\/p>\n
The organic keyword results were then sorted in descending order of volume (V) to identify the phrases that represented the largest numbers of monthly searches\u2014a proxy for interest. Results of this initial search can be found in Table 2.<\/p>\n
Relevant URL Identification<\/h3>\n
The top 10 keywords resulting from the keyword prioritization phase were used as seed keywords to conduct a series of Google searches. For each keyword among these 10, a unique Google search in the US database was performed to produce a unique search engine results page (SERP) from which the resulting top 5 organic page URLs were recorded. This process produced a set of 33 unique URLs.<\/p>\n
Book Title Discovery<\/h3>\n
The 33 unique URLs were then manually inspected to assess relevancy for containing book titles. Of the 33 URLs discovered only 16 were deemed relevant to book title discovery. URLs were disqualified via a manual inspection process on the grounds of either being relevant to only a single book title or not mentioning book titles. This represented a ~49% relevancy rate among URL inclusions.<\/p>\n
Relevant URLs were then manually inspected for book titles and any titles found included were added to a global list of unique titles, up to a maximum of 10 titles per URL source<\/strong>. This upper limit was put into place after observing several URLs to contain excessive numbers of relative titles. Where titles were found to be uniquely identified only by edition, all but the most recent edition were excluded. Another series of filtering for uniqueness was done by ISBN comparison when that data was available. Amazon.com was used as the source for ISBN listings.<\/p>\n Given the statistical nature of this study, it was deemed appropriate to include as many recommendations from as many sources as practical. As such, many titles that were regarded as possibly irrelevant were included for consideration. The frequency of occurrence among several sources was maintained as an effective filter for excluding such titles.<\/p>\n Titles without an ISBN number were excluded. In such cases, these entries were considered non-entries and an additional entry was considered from the list where available. There was only one instance in which this exclusionary filter was applied to a book titled Monroe Doctrine II<\/em>. This book was categorized as fiction.<\/p>\n In the case of Amazon best-seller category lists, some titles appeared multiple times within the top 10 results as multiple formats. For example, the title “Life 3.0: Being Human in the Age of Artificial Intelligence” appeared both in the Kindle edition and Paperback edition. In this case, as in the case of similar duplicates, the physical version was selected and an additional entry was taken from the list.<\/p>\n An additional series of exclusions were made after the titles were collected and frequency analysis was performed. This exclusion removed book titles appearing on fewer than 3 URLs. This filtering was done after noting that, in addition to 61 titles appearing on single URLs only, there were ample titles appearing on 3 or more sources to compile an adequate list.<\/p>\n The results from the title gathering process were then analyzed for frequency of mention among all considered recommending URLs. The list of book titles and corresponding frequency counts were then sorted in descending order to provide a list of the most frequent titles among all URLs recommending books. This ordered data then served as our basis for the list of best books on machine learning and artificial intelligence.<\/p>\n The total organic keywords after having the positional filter applied (P) and the total estimated monthly traffic of the top 5 organic URLs returned for the key term phrase “best machine learning books” are reported in Table 1.<\/p>\n These results reflect known traffic distribution patterns of Google search results such that higher results receive roughly 35% of total clicks resulting in greater estimated monthly traffic counts.<\/p>\n After the collection of organic keyword data for each of these initial 5 URLs, a set of unique keywords found among all URLs was collected. This set was then sorted based on total monthly volume to, again, serve as a proxy representing cumulative interest. There were no filtering criteria applied for inclusion other than the appearance on any of the 5 URLs.<\/p>\n This collection was then sorted in by monthly search volume in descending order. The keywords reflecting the 10 highest monthly search volumes were then selected as the keyword phrases URL discovery in the book title gathering phase. These results are reported in Table 2 below. A JSON-format file containing the collection of all keywords, their relative counts among the considered URLs, and the monthly volumes are available here for download<\/a>.<\/p>\n Of the unique URLs found within the SERP for these terms, a total of 148 book titles were discovered resulting in 89 unique titles for consideration. Of these 89 titles, only 28 appeared on more than one URL. This number was deemed impractical for representing a general consensus among considered URLs such that only book titles appearing on 3 or more distinct URL lists were included. A breakdown of these results is illustrated in Table 3.<\/p>\nExclusion Criteria<\/h3>\n
Title Frequency Sorting<\/h3>\n
Results<\/h2>\n
<\/a><\/a><\/a>
![\"table](\"https:\/\/www.alpharithms.com\/wp-content\/uploads\/963\/table-4-final-title-ranking-results-1024x576.jpg\" "\\"table-4-final-title-ranking-results\\"")
<\/a>