Description

import pandas as pd

data = pd.read_pickle("PATH TO FILE.pkl")
print(data.head())

Series information (English)

Patent Data

The original patent data is given in "engineering-design-knowledge/1-patent-data/"

For our analysis, we sampled 33,881 patents from USPTO using Patents View such that these are stratified according to the CPC subclasses. 

Patent List

In the dataframe "1-patent-list.pkl", we provide the introductory information about these patents as follows.

Patent Sentences

From each patent in the dataset as described above, we acquire the full-text and process the sentences in these. In the dataframe "2-patent-sentences.pkl", we provide the list of 7,566,829 formatted sentences along with patent ID and lengths in terms of token count.

Patent Knowledge Graphs

From the sentences as described above, we extracted 24,537,587 facts of the form - head entity :: relationship :: tail entity using a method described in our prior work - https://arxiv.org/abs/2307.06985

Combining these facts within a patent would form a patent knowledge graph that we examined in the current work to understand basis of design knowledge. The dataframe "3-patent-knowledge-graphs.pkl" provides individual facts as follows. The sentence ID in each row is same as the one mentioned in the previous dataframe - "2-patent-sentences.pkl".

We combine facts as described above, within a patent, to get a knowledge graph that is used for examination in our current work. The results of our analysis are compiled into dataframes as explained below.

Series information (English)

Linguistic Basis

We analysed the frequencies of entities and relationships in the knowledge graphs populated for each patent in the sample. In the dataframes provided under "engineering-design-knowledge/2-linguistic-basis/", we provide information for 5,015,681 entities, 845,303 relationships, and 165 hierarchical relationships regarding their frequencies and linguistic syntaxes. In our work, we fit the proportions of the syntaxes to a Zipf distribution to visualise these at different percentiles.

Entity Syntaxes

Relation Syntaxes

Hierarchical Relation Syntaxes

Series information (English)

Structural Basis

We did motif analysis on the network structures of the patent knowledge graphs to identfy statistically recurrent 3-node and 4-node patterns that are building blocks for each patent knowledge graph. In the dataframe - "1-patent-motifs.pkl" under "engineering-design-knowledge/3-structural-basis/", we list the network size and the motifs for each patent in the sample.

The pattern # mentioned in the motifs field as above could be references using the images included in the folder "pattern-figures".